advanced network analyses of resting-state …
TRANSCRIPT
ADVANCED NETWORK ANALYSES OF RESTING-STATE FUNCTIONAL
CONNECTIVITY TO PROBE EXECUTIVE FUNCTION IN PARKINSON’S DISEASE
AND MULTIPLE SCLEROSIS
by
Sue-Jin Lin
MSc, National Yang-Ming University, 2008
A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF
THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
in
THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES
(Neuroscience)
THE UNIVERSITY OF BRITISH COLUMBIA
(Vancouver)
August 2018
© Sue-Jin Lin, 2018
ii
The following individuals certify that they have read, and recommend to the Faculty of Graduate and Postdoctoral Studies for acceptance, the dissertation entitled: Advanced network analyses of resting-state functional connectivity to probe executive function in Parkinson’s Disease and Multiple Sclerosis
submitted by Sue-Jin Lin in partial fulfillment of the requirements for
the degree of Doctor of Philosophy
in Neuroscience
Examining Committee:
Martin J. McKeown, Faculty of Medicine
Supervisor
Anthony Traboulsee, Faculty of Medicine
Supervisory Committee Member
Supervisory Committee Member
Luke Clark, Department of Psychology
University Examiner
Teresa Liu-Ambrose, Department of Physical Therapy
University Examiner
Additional Supervisory Committee Members:
Alex MacKay, Department of Physics and Astronomy
Supervisory Committee Member
Jane Wang, Department of Electrical and Computer Engineering Program
Supervisory Committee Member
Supervisory Committee Member
iii
Abstract
This thesis is to probe the systems-level neurobiological bases for executive function in patient
populations overarchingly.
We focus on two representative diseases, Parkinson’s Disease (PD), and Multiple Sclerosis (MS),
as although they have different pathologies, patients often result in similar cognitive deficits. We
examine resting-state fMRI data from both PD and MS subjects with novel methods in a network
fashion. We employ advanced connectivity analyses to evaluate graph theoretical, static and
dynamic resting-state functional connectivity (rsFC) measures. Multivariate statistical methods
such as Canonical Correlate Analysis (CCA) and Multiset Canonical Correlate Analysis (MCCA)
are used to robustly link rsFC and cognitive performance. PD data used in the thesis research
include three cohorts: Parkinson’s Progression Markers Initiative (PPMI) and two research
projects conducted at UBC (project name: BCT and GFM2). For MS, two cohorts are included:
OPERA MS clinical trial and COGMS research project, which are both conducted at UBC.
After a general introduction in the first chapter, in the second chapter we examine multivariate
relations between demographic and cognitive profiles with CCA, showing that female gender is
associated with better cognitive performance in both diseases possibly due to protective effects of
estrogen.
In chapter 3, we use correlation to assess functional connections. Both diseases have significantly
altered interhemispheric connectivity, which is associated with altered cognitive performance in
MS, but not PD.
In chapter 4, we utilize graph theoretical approaches and find increased segregation of rsFC in PD,
supporting a previously-proposed model of vulnerability of hubs in disease populations. In MS,
higher modularity of the rsFC network is correlated with better executive skills.
iv
In chapter 5, we explore dynamic rsFC and discover that longer disease duration in MS is
associated with decreased dynamic rsFC. In both populations, dynamic interhemispheric
connectivity is robustly associated with cognitive abilities.
In chapter 6, MCCA is applied to jointly explore the associations between dynamic and stationary
rsFC, and behavioural measures. In MS, better executive functioning is supported by higher
education, stronger and dynamic rsFC; in PD, better memory function is related to segregated brain
networks and dynamics of interhemispheric connections.
Chapter 7 summarizes and concludes these chapters.
v
Lay Summary
Cognitive deficits are a very troubling symptom for people with neurological disease.
Understanding cognitive deficits has been difficult because there does not seem to be a simple
relationship between damages to one part of the brain and the deficits. It seems that disease effects
over widespread brain areas are more associated with cognitive deficits. In this research, we
investigated functional connectivity (FC), i.e. how brain regions communicate through
information transfer, and its relations to cognitive deficits in two diseases showing similar
cognitive impairments: Parkinson’s Disease and Multiple Sclerosis. Novel analyses with
functional magnetic resonance imaging (fMRI) data were used to examine FC and advanced
statistical methods were used to link FC and cognitive function. In both diseases, we found robust
associations between network level descriptions of FC and performance on cognitive tests.
Determining the networks associated with cognitive performance is a first step towards targeted
therapy attempting to reduce deficits.
vi
Preface
The thesis work was conducted in the Pacific Parkinson’s Research Centre at UBC Hospital with
collaboration of the UBC MRI Research Centre and Multiple Sclerosis Clinic at UBC Hospital.
For the study design, I contributed to parts of decision such as resting-state fMRI and cognitive
tests in order to identify ways to approach the problem. For other parts of the research, I helped
with data entry and contributed to annual reports of clinical trials. With some technical help from
colleagues and the supervisor, I performed all the analyses that are included in the thesis chapters
as well as interpreted the results. Several ethics approvals were issued by the UBC Research Ethics
Board (H12-01510, H09-02016, H04-70177, H13-01230).
Chapter 1 is based on a review article written by me with help from Drs. Katrina McMullen,
Pratibha Surathi, Silke Appel-Cresswell, and Martin J. McKeown. The manuscript is ready for
submission.
Chapter 2 includes data from the Parkinson’s Progression Markers Initiative (PPMI) and an
operating grant from the Multiple Sclerosis Society of Canada (to Drs. Alex MacKay and Martin
J. McKeown). PPMI – a public-private partnership – is funded by the Michael J. Fox Foundation
for Parkinson’s Research and funding partners, including AbbVie, Allergan, Avid, Biogen,
BioLegend, Bristol-Myers Squibb, GE Healthcare, Genentech, GlaxoSmithKline, Lilly,
Lundbeck, Merck, Meso Scale Discovery, Pfizer Inc, Piramal Imaging, F. Hoffmann-La Roche
AG, Sanofi Genzyme, Servier, Takeda, Teva, UCB, and Global Capital. A version of chapter 2
(Multiple Sclerosis (MS) data) has been published. Lin SJ, Lam J, Beveridge S, Vavasour I,
Traboulsee A, Li DKB, MacKay A, McKeown M, Kosaka B. (2017) Cognitive Performance in
vii
Subjects With Multiple Sclerosis Is Robustly Influenced by Gender in Canonical-Correlation
Analysis. J Neuropsychiatry Clin Neurosci. 29(2):119-127. The Parkinson’s Disease (PD) data in
chapter 2 has been included as part of the publication. Lin SJ, Baumeister TR, Gard S, McKeown
M. (2018) Cognitive Profiles and Hub Vulnerability in Parkinson’s Disease. Front Neurol,
2018;9:1-13.
Chapter 3 includes data from the Pacific Parkinson’s Research Centre (PPRC) at UBC Hospital
and a Phase III clinical trial of MS (OPERA II; NCT01412333) conducted in the MS Clinic at
UBC Hospital and supported by F. Hoffmann-La Roche, Ltd. A version of chapter 3 (MS data)
has been submitted. I wrote the manuscript and performed analyses with help from coauthors:
Kolind S, Liu A, McMullen K, Vavasour I, Wang ZJ, Traboulsee A, McKeown M.
Chapter 4 includes data from PPMI and an operating grant from the Multiple Sclerosis Society of
Canada (to Drs. Alex MacKay and Martin J. McKeown) as chapter 2. A version of chapter 4
(control data) has been published as a poster. Lin SJ, Baumeister TR, MacKay A, Vavasour I, Li
DKB, McKeown M. (2017) Reproducibility of graphical measures and dynamic network features
in resting state fMRI. 25th Annual Meeting of the International Society for Magnetic Resonance in
Medicine. The PD data in chapter 4 has been included as part of the publication. Lin SJ, Baumeister
TR, Gard S, McKeown M. (2018) Cognitive Profiles and Hub Vulnerability in Parkinson’s
Disease. Front Neurol, 2018;9:1-13.
Chapter 5 includes data from an operating grant from the Multiple Sclerosis Society of Canada (to
Drs. Alex MacKay and Martin J. McKeown) and a project in PPRC conducted by Laura Kim and
colleagues. A version of chapter 5 (control data) has been published as a poster. Lin SJ, Baumeister
TR, MacKay A, Vavasour I, Li DKB, McKeown M. (2017) Reproducibility of graphical measures
viii
and dynamic network features in resting state fMRI. 25th Annual Meeting of the International
Society for Magnetic Resonance in Medicine. The MS data in chapter 5 has been included as part
of the accepted manuscript. Lin SJ, Vavasour I, Kosaka B, Li DKB, Traboulsee A, MacKay A,
McKeown MJ. Education, and the Balance Between Dynamic and Stationary Functional
Connectivity Jointly Support Executive Functions in Relapsing-Remitting Multiple Sclerosis.
Human Brain Mapp, 2018 (Accepted).
Chapter 6 includes data from an operating grant from the Multiple Sclerosis Society of Canada (to
Drs. Alex MacKay and Martin J. McKeown) and a project in PPRC conducted by Laura Kim and
colleagues. The MS data in chapter 6 has been included as part of the accepted manuscript as
chapter 5. The PD data in chapter 6 has been included in a preparation of journal submission.
ix
Table of Contents
Abstract ......................................................................................................................................... iii
Lay Summary .................................................................................................................................v
Preface ........................................................................................................................................... vi
Table of Contents ......................................................................................................................... ix
List of Tables ............................................................................................................................. xvii
List of Figures ........................................................................................................................... xviii
List of Abbreviations ...................................................................................................................xx
Acknowledgements .................................................................................................................... xxi
Dedication .................................................................................................................................. xxii
Chapter 1: Introduction ................................................................................................................1
1.1 Executive dysfunction and neuroimaging in Multiple Sclerosis and Parkinson’s Disease ........ 2
1.1.1 Multiple Sclerosis ....................................................................................................... 2
1.1.2 Parkinson’s Disease .................................................................................................... 5
1.1.3 Summary ..................................................................................................................... 6
1.2 Functional connectivity and analysis approaches in clinical research ............................ 7
1.2.1 Resting-state functional connectivity .......................................................................... 7
1.2.1.1 Common analyses and applications .................................................................... 9
1.2.2 Dynamic functional connectivity .............................................................................. 10
1.2.2.1 Methods and applications ................................................................................. 11
1.3 Resting-state functional connectivity and executive function in Multiple Sclerosis and
Parkinson’s Disease .................................................................................................................. 13
x
1.3.1 Multiple Sclerosis ..................................................................................................... 13
1.3.2 Parkinson’s Disease .................................................................................................. 16
1.3.3 Summary ................................................................................................................... 25
1.4 Impacts and contributions of studies on resting-state functional connectivity to clinical
neuroscience .............................................................................................................................. 25
1.4.1 Key connections in cognitive deficits ....................................................................... 25
1.4.2 Three components of executive function .................................................................. 27
1.4.3 Unique features of brain organization in different neurological conditions may all
result in executive dysfunction ............................................................................................. 31
1.5 Research objectives and thesis organization ................................................................. 32
Chapter 2: Cognitive profiles in Parkinson’s Disease and Multiple Sclerosis .......................36
2.1 Introduction of cognitive profiles in Parkinson’s Disease and Multiple Sclerosis ....... 36
2.2 Materials and methods .................................................................................................. 40
2.2.1 Subjects ..................................................................................................................... 40
2.2.2 Test battery................................................................................................................ 43
2.2.3 Analysis..................................................................................................................... 45
2.3 Results ........................................................................................................................... 47
2.3.1 PD ............................................................................................................................. 47
2.3.2 MS ............................................................................................................................. 48
2.4 Discussion ..................................................................................................................... 52
2.4.1 Cognitive profiles in PD: female gender and education support better cognitive
function, while comorbidity is related to poor cognition ...................................................... 52
xi
2.4.2 Cognitive profiles in MS: female gender supports better cognitive function and
disease severity is related to worse performance .................................................................. 55
2.4.3 Commonalities and differences ................................................................................. 57
2.4.4 Limitations ................................................................................................................ 59
2.5 Conclusion .................................................................................................................... 59
Chapter 3: Resting-state functional connections and cognition ..............................................61
3.1 Introduction ................................................................................................................... 61
3.1.1 Effects of interhemispheric and long-range connections to cognition ...................... 61
3.1.2 Interhemispheric and long-range connections in Parkinson’s Disease and Multiple
Sclerosis ................................................................................................................................ 63
3.1.3 Analyses .................................................................................................................... 64
3.2 Materials and methods .................................................................................................. 65
3.2.1 Subjects ..................................................................................................................... 65
3.2.2 Image acquisition ...................................................................................................... 67
3.2.3 Image preprocessing ................................................................................................. 67
3.2.4 Functional connectivity analysis ............................................................................... 69
3.3 Results ........................................................................................................................... 72
3.4 Discussion ..................................................................................................................... 79
3.4.1 Enhanced interhemispheric connectivity may indicate compensatory mechanisms ........... 79
3.4.2 Interhemispheric connectivity predicts cognitive performance ................................ 81
3.4.3 Sensitivity of interhemispheric connectivity to different cognitive tests .................. 83
3.4.4 Altered long-range connections ................................................................................ 83
3.5 Limitation ...................................................................................................................... 84
xii
3.6 Conclusion .................................................................................................................... 85
Chapter 4: Brain organization and executive function ............................................................87
4.1 Introduction ................................................................................................................... 87
4.1.1 Connections and networks of rsFC and the relations to cognition ........................... 87
4.1.2 Graphical measures and cognition ............................................................................ 88
4.1.3 Hub structures and cognition .................................................................................... 89
4.1.4 Study aims ................................................................................................................. 90
4.2 Materials and methods .................................................................................................. 91
4.2.1 Subjects and behavioural data ................................................................................... 91
4.2.2 Imaging acquisition ................................................................................................... 94
4.2.3 Preprocessing ............................................................................................................ 95
4.2.4 Functional connectivity analysis ............................................................................... 96
4.2.5 Brain-behaviour analysis .......................................................................................... 99
4.3 Results ......................................................................................................................... 100
4.3.1 Reproducibility ....................................................................................................... 100
4.3.2 Parkinson’s Disease ................................................................................................ 101
4.3.2.1 Graphical measures ......................................................................................... 101
4.3.2.2 Brain-behaviour association............................................................................ 105
4.3.3 Multiple Sclerosis ................................................................................................... 106
4.3.3.1 Graphical measures ......................................................................................... 106
4.3.3.2 Brain-behaviour association............................................................................ 106
4.4 Discussion ................................................................................................................... 111
4.4.1 Changes of brain organization ................................................................................ 111
xiii
4.4.2 Reduced functional integration and increased functional segregation correlated with
better cognitive performance .............................................................................................. 114
4.4.3 Predictability of graphical measures to behaviour .................................................. 118
4.4.4 Limitations .............................................................................................................. 120
4.5 Conclusion .................................................................................................................. 120
Chapter 5: Dynamic functional connectivity and executive function ...................................122
5.1 Introduction ................................................................................................................. 122
5.1.1 Dynamic functional connectivity ............................................................................ 122
5.1.2 Dynamic functional connectivity and cognition ..................................................... 123
5.1.3 Dynamic functional connectivity in diseased populations ...................................... 124
5.2 Materials and methods ................................................................................................ 126
5.2.1 Subjects and behavioural data ................................................................................. 126
5.2.2 Imaging acquisition ................................................................................................. 128
5.2.3 Preprocessing .......................................................................................................... 128
5.2.4 Dynamic functional connectivity analysis .............................................................. 130
5.2.5 Correlation and regression analyses........................................................................ 136
5.3 Results ......................................................................................................................... 137
5.3.1 Reproducibility ....................................................................................................... 137
5.3.2 Multiple Sclerosis ................................................................................................... 138
5.3.2.1 Feature comparison ......................................................................................... 138
5.3.2.2 Correlation and regression .............................................................................. 139
5.3.3 Parkinson’s Disease ................................................................................................ 143
5.3.3.1 Feature comparison ......................................................................................... 143
xiv
5.3.3.2 Correlation and regression .............................................................................. 144
5.4 Discussion ................................................................................................................... 147
5.4.1 Changes in dynamic functional connectivity may indicate compensation in MS .. 147
5.4.2 Disease duration is robustly modulated by dynamic functional features in MS..... 149
5.4.3 Disease severity can be predicted by dynamic features in PD ................................ 151
5.4.4 Memory function is robustly associated with dynamics in PD .............................. 152
5.4.5 Limitations .............................................................................................................. 153
5.5 Conclusion .................................................................................................................. 154
Chapter 6: Dynamic brain organization, disease severity, and executive function .............155
6.1 Introduction ................................................................................................................. 155
6.1.1 Stationary and dynamic functional connectivity ..................................................... 155
6.1.1.1 Cognition and stationary functional connectivity: graph theoretical analysis 156
6.1.1.2 Cognition and dynamic functional connectivity: time-varying approach ....... 157
6.1.2 Studies looking at both stationary and dynamic aspects of functional connectivity ......... 158
6.1.3 Study aims ............................................................................................................... 159
6.2 Materials and methods ................................................................................................ 160
6.2.1 Subjects ................................................................................................................... 160
6.2.2 Neuropsychological and clinical assessments ........................................................ 161
6.2.3 Imaging acquisition and processing ........................................................................ 162
6.2.4 Connectivity analysis .............................................................................................. 163
6.2.5 Brain-behaviour associations .................................................................................. 164
6.3 Results ......................................................................................................................... 165
6.3.1 MCCA in MS .......................................................................................................... 165
xv
6.3.2 MCCA in PD........................................................................................................... 169
6.4 Discussion ................................................................................................................... 172
6.4.1 Relations between dynamic, stationary functional connectivity, education, and
cognition in MS................................................................................................................... 172
6.4.2 Disease effects on cognition and comorbidity in MS ............................................. 173
6.4.3 Better memory function in PD is related to dynamic interhemispheric connectivity
and stationary network segregation .................................................................................... 174
6.4.4 Disease severity and comorbidities are related to compensated dynamics in PD .. 176
6.4.5 Limitations and future work.................................................................................... 177
6.5 Conclusion .................................................................................................................. 178
Chapter 7: Conclusion ...............................................................................................................180
7.1 Summary of the research ............................................................................................ 180
7.1.1 Cognitive impairments and functional connectivity ............................................... 181
7.1.2 Cognitive profiles and the associations to demographical features ........................ 181
7.1.3 Long-range connections, interhemispheric connectivity, and cognitive function .. 182
7.1.4 Topological brain organization and executive function .......................................... 183
7.1.5 Dynamic functional connectivity and executive function ...................................... 185
7.1.6 Combining dynamic, stationary functional connectivity and behavioural data reveals
complementary information ................................................................................................ 186
7.2 Strengths and limitations............................................................................................. 188
7.2.1 Strengths: functional connectivity analysis ............................................................ 188
7.2.2 Strengths: tested domains ....................................................................................... 189
7.2.3 Strengths: brain-behaviour analysis ........................................................................ 190
xvi
7.2.4 Limitations: study design ........................................................................................ 190
7.2.5 Limitations: gender issues....................................................................................... 191
7.2.6 Limitations: compensatory effects .......................................................................... 192
7.3 Future research directions ........................................................................................... 193
Bibliography ...............................................................................................................................195
Appendices ..................................................................................................................................229
Appendix A ............................................................................................................................. 229
A.1 Canonical loadings of individual PD subjects. ....................................................... 229
Appendix B ............................................................................................................................. 230
B.1 Robust regression in PD on-medication ................................................................. 230
B.2 Whole brain connectivity matrix of HC and PD ..................................................... 231
B.3 Correlation between long-range connections and cognitive performance .............. 231
Appendix C ............................................................................................................................. 233
C.1 Significantly different local measures between PD and HS ................................... 233
C.2 Logistic LASSO identifies graphical measures into PD and HS groups ................ 234
C.3 Canonical correlation analysis (CCA) between cognitive performance and global
measures .............................................................................................................................. 236
Appendix D ............................................................................................................................. 239
D.1 Linear regression model of dynamic features and WCSTCC ................................. 239
D.2 The component which significantly modulated cognitive flexibility ..................... 240
xvii
List of Tables
Table 1.1 Summary of the rsFC studies in MS and PD ................................................................ 23
Table 2.1 Demographics, clinical data, and cognitive scores in PD subjects ............................... 42
Table 2.2 Demographics, clinical measures, and neurological scores in MS subjects ................. 43
Table 3.1 Demographics of MS and HC subjects ......................................................................... 66
Table 3.2 Demographics of PD and HC subjects ......................................................................... 67
Table 3.3 The 38 ROIs in the study for MS (19 bilateral regions) ............................................... 69
Table 3.4 The 54 ROIs in the study for PD (27 bilateral regions) ................................................ 69
Table 4.1 Demographics, clinical data, and cognitive scores in PD and HS ................................ 93
Table 4.2 Demographics in MS and NC ....................................................................................... 93
Table 4.3 68 ROIs are used in the PD and MS studies ................................................................. 97
Table 4.4 Definitions of the graphical measures used in this study……………………………...97
Table 4.5 COV of all graphical measures across fMRI sessions ................................................ 101
Table 4.6 Regression results of local measures in MS ............................................................... 110
Table 5.1 Clinical, demographical, and cognitive assessment scores in PD and NC ................. 128
Table 5.2 Eighteen bilateral regions-of-interest (ROIs) in the connectivity analysis in MS study
and healthy subjects for reproducibility ...................................................................................... 129
Table 5.3 Mathematical definitions of network features learned in the sliding window approach ....... 135
Table 5.4 COV results of dynamic features across 3 fMRI sessions .......................................... 138
Table 5.5 Significant linear regression model in MS ................................................................. 143
Table 5.6 Two significant linear regression modes in PD .......................................................... 146
Table 7.1 The summary of data used in each chapter ................................................................. 180
xviii
List of Figures
Figure 1.1 The connections and regions of rsFC correlated with executive performance in HS,
MS, and PD ................................................................................................................................... 24
Figure 1.2 A model of executive function .................................................................................... 29
Figure 1.3 A diagram of the flow in this thesis research .............................................................. 33
Figure 2.1 The CCA model shows inter-correlated cognitive patterns with demographics and
clinical data in PD ......................................................................................................................... 49
Figure 2.2 CCA results of timed cognitive tests in MS ................................................................ 49
Figure 2.3 CCA results of untimed cognitive tests in MS ............................................................ 51
Figure 3.1 The procedure of regression analysis .......................................................................... 71
Figure 3.2 Functional interhemispheric connectivity in normal controls and MS ....................... 73
Figure 3.3 Functional interhemispheric connectivity in healthy controls, PF on-medication, and
PD off-medication ......................................................................................................................... 74
Figure 3.4 Relationships between real scores and predicted scores in SDMT and PASAT tests
based on cross-validation .............................................................................................................. 75
Figure 3.5 Important brain regions for cognitive performances in MS ........................................ 76
Figure 3.6 Connections that are significantly different between HC and MS .............................. 78
Figure 3.7 Connections that are significantly different between PD off-medication and PD on-
medication ..................................................................................................................................... 78
Figure 4.1 rmANOVA results of graphical measures across 3 fMRI sessions ........................... 100
Figure 4.2 Local graphical measures in PD ................................................................................ 103
Figure 4.3 Local graphical measures in HS ................................................................................ 104
xix
Figure 4.4 Spearman’s correlation between cognitive scores and graphical measures in PD .... 105
Figure 4.5 Spearman’s correlation between cognitive scores and graphical measures in MS ... 107
Figure 4.6 Regression results of local measures in MS .............................................................. 110
Figure 4.7 Prediction results of local measures in MS ............................................................... 111
Figure 5.1 The sliding window approach with Pearson’s correlation ........................................ 132
Figure 5.2 The sliding window approach with inverse covariance matrix ................................. 134
Figure 5.3 rmANOVA results of dynamic features across 3 fMRI sessions .............................. 137
Figure 5.4 Results of dynamic feature comparison between MS and NC .................................. 139
Figure 5.5 Significant correlations between disease duration and dynamic features in MS ...... 141
Figure 5.6 PCA results of dynamic features in MS .................................................................... 142
Figure 5.7 The adjusted variable plot of the linear regression model of PCA components and
disease duration in MS ................................................................................................................ 143
Figure 5.8 Correlation between FOCcs and MoCA sub-score DELY in PD ............................. 144
Figure 5.9 Four principal components of the dynamic features in PD ....................................... 145
Figure 5.10 The adjusted variable plot of linear regression modes in PD .................................. 147
Figure 6.1 Correlation between MCCA sets in MS .................................................................... 167
Figure 6.2 MCCA loadings in MS .............................................................................................. 168
Figure 6.3 Correlation coefficients between 4 MCCA data sets in PD ...................................... 170
Figure 6.4 MCCA loadings in PD .............................................................................................. 171
xx
List of Abbreviations
rsfMRI resting-state functional magnetic resonance imaging
MS multiple sclerosis
PD Parkinson’s disease
rsFC resting-state functional connectivity
sMRI structural magnetic resonance imaging
BOLD blood oxygenation level dependent
RSNs resting-state networks
MCI mild cognitive impairment
UPDRS Unified Parkinson's Disease Rating Scale
EDSS Expanded Disability Status Scale
MMSE Mini Mental State Examination
MoCA Montreal Cognitive Assessment
SDMT Symbol Digit Modalities Test
PASAT Paced Auditory Serial Addition Test
rmANOVA repeated measure ANOVA
COV coefficient of variation
dFC dynamic functional connectivity
drsFC dynamic resting-state functional connectivity
PCA principal component analysis
CCA canonical correlation analysis
MCCA multiset canonical correlation analysis
xxi
Acknowledgements
I thank my supervisor Dr. Martin McKeown and committee members Drs. Alex MacKay, Anthony
Traboulsee, and Jane Wang as well as external committee Dr. Silke Appel-Cresswell for their
feedback and suggestions through the training. I also acknowledge funding support from the
Ministry of Education, Taiwan and F. Hoffmann-La Roche, Ltd. Several projects included in the
thesis are conducted with help from colleagues in the Pacific Parkinson’s Research Centre,
Multiple Sclerosis clinic trial, Psychiatry Department, and UBC MRI Research Centre at UBC. I
thank everyone who is involved in the research projects. I also thank Tobias R Baumeister and
Saurabh Garg for their technical help with data analysis as well as Aiping Liu, Jessie Fu, and Jiayue
Cai for their contributions to processing scripts.
In addition, I am very grateful for having friends who support each other in the Taiwanese
Graduate Student Association in Vancouver, especially my former housemates Ting-Chun Kuo
and Yi-Chen Lee. It is important to have them as we are all international students who come to
Canada without families, and they understand the difficulties and challenges that we face every
day. Their friendship and support have become one of the most important things in my life here.
Finally, there are no words that can express my gratitude to Roberto, who supported me in many
ways every single day in the past 5 years. Without him, I would not be able to finish the work
without developing severe illness. The closest language that I can use to show my appreciation is
his mother tongue: Vielen Dank. Ich bin froh, dass ich dich in Kanada getroffen habe.
xxii
Dedication
To myself
給這幾年經歷風霜的自己: 終於熬過來了
1
Chapter 1: Introduction
This chapter provides a general introduction of the cognitive deficits, related studies on functional
connectivity, and common analyses in two neurological conditions: Parkinson’s Disease and
Multiple Sclerosis. In addition, the summary relates findings of previous research to a proposed
model of higher-order cognitive function. Finally, suggested approaches to study the neuronal
bases of cognitive function are made.
Executive function is a set of higher-order cognitive processes that include inhibition, working
memory, cognitive flexibility, reasoning, problem-solving, and planning to achieve goal-directed
behaviour. Executive functions are important for activities of daily living and a satisfactory quality
of life [Diamond, 2013], and poor executive function leads to decreased productivity and problem
solving abilities at work and school [Bailey, 2007; Blair and Peters Razza, 2007; Brown and
Landgraf, 2010]. As a variety of neurological disorders can result in executive dysfunction,
understanding the mechanisms of these deficits has become necessary in order to guide therapeutic
options [Elliott, 2003]. Attempts at clinicopathological correlation, where damage to specific
regions is correlated with particular deficits, were modestly successful at best with cognitive
deficits, and this has been referred to as the clinical-radiological paradox [Barkhof, 2002].
Traditional clinical imaging focusing on damage to particular brain regions may obscure more
diffuse structural and functional connectivity changes which are likely more important for complex
cognitive processes such as executive function. To overcome this, a combination of novel brain
imaging techniques such as resting-state functional magnetic resonance imaging (rsfMRI) and
suitable analyses such as network measures and multivariate approaches can be used to investigate
2
correlations between different behaviour and associated connectivity patterns [Hackmack et al.,
2012; van den Heuvel and Hulshoff Pol, 2010; Smith et al., 2009].
Executive dysfunction is prominent in several neurological and psychiatric illnesses including
frontal lobe damage, basal ganglia disorders, depression and schizophrenia [Elliott, 2003]. This
dissertation focuses on two neurological disorders known to impact executive function: Multiple
Sclerosis (MS) and Parkinson’s Disease (PD). These diseases have been chosen because they
represent entirely different primary pathologies (myelin damage vs basal ganglia dysfunction),
affect different ages and genders, yet still result in executive dysfunction. Contrasting and
comparing the cognitive deficits in these two diseases are expected to provide insights into the
common origins of executive dysfunction. We focus especially on resting-state functional
connectivity (rsFC) in these two neurological conditions, as this will examine distributed
representations that are complementary to the traditional clinicopathological correlation model.
1.1 Executive dysfunction and neuroimaging in Multiple Sclerosis and Parkinson’s
Disease
These two conditions are well-described progressive diseases of the central nervous system, which
result in executive dysfunction. MS is characterized by white matter lesions and PD is a basal
ganglia disorder with dopamine depletion. This section reviews the executive deficits in MS and
PD as well as the relations to traditional neuroimaging measures.
1.1.1 Multiple Sclerosis
MS is a neuroinflammatory disease which shows widespread lesions (demyelinated axons attacked
by T cells) mainly in the white matter and spinal cord but occasionally in the grey matter, resulting
3
in a broad range of deficits such as motor, cognitive, and neuropsychiatric symptoms [Goldenberg,
2012]. Four subtypes of MS have been defined: relapsing-remitting, secondary progressive,
primary progressive and progressive-relapsing MS depending upon the pattern of temporal
evolution of symptomatology. This review focuses on the most common subtype of MS –
relapsing-remitting MS (RRMS).
In RRMS, up to 70% of patients demonstrate cognitive impairment, with the most commonly
affected domains being information processing speed, attention, visuospatial ability, memory, and
executive function [Chiaravalloti and DeLuca, 2008; Langdon, 2011; Wallin et al., 2006]. The
affected domain incorporating attention involves complex processes such as alertness,
selective/focused/divided attention, and vigilance rather than “simple” attention (e.g. as tested by
asking a subject to repeat a series of digits) [Chiaravalloti and DeLuca, 2008; Guimarães and Sá,
2012]. Processing speed, a commonly impaired domain in MS, is required in many cognitive tasks
to rapidly react to processed information [Chiaravalloti and DeLuca, 2008; Guimarães and Sá,
2012]. Research has suggested that difficulty in acquiring new knowledge in MS might be a greater
problem than information retrieval in the memory domain [Chiaravalloti and DeLuca, 2008].
Executive dysfunction in MS also includes poor planning, decreased working memory, set-shifting
and mental flexibility problems [Foong et al., 1997; Holland et al., 2014]. Patients with MS
generate significantly fewer words in the Verbal Fluency Test, take longer time to complete the
Stroop test, make more errors in the Stroop test and spatial working memory test, and require
longer time to initiate motor execution in a test of planning and problem solving [Foong et al.,
1997]. In a young MS population, 35% of the subjects score lower than the normative data on the
Trail Making Test B (TMT B), suggesting disabilities of mental flexibility [Holland et al., 2014].
4
Among the cognitive deficits, executive dysfunction seems to impact patients’ quality of life and
occupational performance the most [Preston et al., 2013].
Traditional MRI methods have been explored in MS as biomarkers for cognitive impairment, but
results have only met with modest success. Diffuse lesions observed in fluid-
attenuated inversion recovery (FLAIR) MRI sequences and T2-weighted images in the
periventricular white matter and the commissural fiber tracts are associated with processing speed
deficits, working memory difficulties, perception and spatial processing [Houtchens et al., 2007;
Rossi et al., 2012; Stankiewicz et al., 2009]. However, only a few studies have actually shown
strong correlations between executive deficits and particular MRI-detectable lesion locations in
MS [Filippi et al., 2010]. This has raised concerns whether lesion location (or even overall lesion
load) can be a specific indicator of cognitive deficits, especially executive dysfunction [Rocca et
al., 2014]. It is this observation that has motivated a network approach to cognitive deficits as done
in this thesis.
Newer imaging methods have proved more successful in finding a link between imaging features
and cognitive dysfunction in MS. Decreased white matter integrity, as evaluated by fractional
anisotropy (FA)-based graphical measures in the frontoparietal network, subcortical regions, and
insula are associated with impaired attention and executive performance [Llufriu et al., 2017].
Cortical thinning in the left anterior cingulate, superior frontal, lateral orbitofrontal, and superior
parietal regions have been shown to be associated with executive deficits (e.g. verbal fluency)
[Geisseler et al., 2016].
5
1.1.2 Parkinson’s Disease
PD is a neurodegenerative movement disorder that is characterized by degeneration of
dopaminergic neurons in the substantia nigra pars compacta [Dauer and Przedborski, 2003],
affecting dopaminergic pathways such as nigrostriatal and mesocortical pathways [Ouchi et al.,
2001]. In addition to dopamine depletion, several factors have been considered contributing to the
impairments in PD such as protein misfolding, impaired cholinergic activity, and adenosine
receptor abnormalities [Aarsland et al., 2017]. This dissertation focuses on sporadic PD (i.e. there
is no apparent genetic linkage), which is the most common type [Dauer and Przedborski, 2003].
Executive dysfunction has been commonly reported, using tasks which require planning, set-
shifting, control of attention, working memory, and timing skills such as the Wisconsin Card
Sorting Test (WCST), TMT, Tower of London Test (ToL), and time perception task [Dirnberger
and Jahanshahi, 2013; Goldman and Litvan, 2011; Palavra et al., 2013; Parker et al., 2013;
Williams-Gray et al., 2007]. In the dual syndrome hypothesis, the frontal cognitive subtype, more
commonly seen in tremor-predominant PD, is associated with decreased dopamine levels in
frontal-striatal circuits, resulting in impaired executive function such as reduced verbal fluency
and impaired planning ability [Kehagia et al., 2010; Siepel et al., 2014; Williams-Gray et al., 2007].
The posterior cognitive subtype, associated with a more profound cholinergic deficit, demonstrates
impaired visuospatial abilities, and dementia [Miller et al., 2013]. Choline acetyltransferase
(ChAT) activity can be reduced in the hippocampus, prefrontal cortex, and temporal cortex and
the degradation is correlated with global cognitive decline in PD [Mattila et al., 2001]. Positron
emission tomography (PET) studies have shown that cholinergic degradation in PD correlates with
performance of attention, working memory, and executive tests, while this degradation is not
6
correlated with motor symptoms [Bohnen et al., 2006]. Thus both dopamine and cholinergic
signaling have been shown to affect cognitive domains such as working memory, decision making,
and attention [Ballinger et al., 2016; Takahashi et al., 2012]. In fact, decreased cholinergic and
dopaminergic signaling and neuronal death may be joint pathological features in PD which affect
cognition [Ballinger et al., 2016].
Other pathological changes can also be observed and linked to executive impairment in PD.
Cortical thinning in the dorsolateral superior frontal gyrus as well as white matter abnormalities
underlying bilateral frontal and temporal cortices have demonstrated relationships to executive
dysfunction [Koshimori et al., 2016]. Diffusion tensor imaging (DTI) has demonstrated a positive
correlation between FA in frontal-subcortical regions and executive scores [Gallagher et al., 2013];
likewise, executive function has been positively correlated with FA and negatively correlated with
mean diffusivity (MD) in frontal white matter tracks such as the anterior limb of the internal
capsule and the genu of the corpus callosum [Zheng et al., 2014]. Taken together, structural
changes in the frontal areas are associated with executive dysfunction in PD as well as some
temporal/posterior regions. Neurotransmitter abnormalities are also related to cognitive
impairments in attention, memory, and executive domains.
1.1.3 Summary
MS and PD are two distinct neurological conditions, which both present with executive
dysfunction regardless of the location of lesions and affected areas/circuits, supporting the notion
that executive function requires widely distributed regions to coordinate as a whole rather than
engaging one specific region of the cerebral cortex such as the prefrontal cortex [Alvarez and
Emory, 2006; Elliott, 2003; Heyes, 2012; McIntosh, 2000; McIntosh, 2004; Miller and Wallis,
7
2009]. Newer techniques such as fMRI allow observing neural events with a broader outlook and
examining brain regions that are active under certain conditions. When these techniques are used
to look at the temporal co-activation of regions they move beyond a simple localization approach
(which region does what function) to looking at distributed networks of activation (which
regions are acting in concert or sequence with each other). This approach is referred to as
functional connectivity, which allows investigations of how anatomically segregated brain regions
communicate [Elliott, 2003; van den Heuvel and Hulshoff Pol, 2010].
1.2 Functional connectivity and analysis approaches in clinical research
1.2.1 Resting-state functional connectivity
Neuroimaging techniques have been applied to study executive function in order to understand its
neurobiological correlates [Bunge and Souza, 2009; Chung et al., 2014]. With structural magnetic
resonance imaging (sMRI), structural changes such as grey matter atrophy and altered cortical
thickness in brain regions have been compared with the performance of executive tasks that are
assessed outside of the MRI scanner. With functional modalities, such as fMRI and PET, subjects
perform executive tasks in the scanner while the functional images are being acquired to monitor
neuronal activation patterns and metabolism triggered by the cognitive demands of the tasks.
Recently, rsfMRI has been used rather than, or in conjunction with, tasked-driven fMRI to study
brain activity. rsfMRI involves scanning participants who are awake but not engaged in direct
cognitive action (subjects are asked to not think about anything in particular, also known as
spontaneous or non-directed cognition) and may be more appropriate for disease populations
where poor behavioural performance in task based studies may affect interpretation of studies.
8
Analysis of task-driven fMRI studies is usually based on testing hypothesized amplitude-related
blood oxygenation level dependent (BOLD) changes at different brain regions. With rsfMRI one
does not have a priori hypothesized waveform as participants are not performing specific tasks
and data driven methods, such as assessing rsFC are used for analysis. rsFC describes the statistical
association of neural patterns between distinct brain regions during rest [Friston, 1994; van den
Heuvel and Hulshoff Pol, 2010]. This ability to apply an exploratory approach is advantageous
when considering compensatory and adaptive mechanisms as there is little a priori knowledge of
which brain areas may demonstrate such behaviour. Although subjects do not perform specific
cognitive tasks during rsfMRI studies, patterns of rsFC have been linked to performance on
subsequent cognitive tasks [Diez et al., 2015; van den Heuvel and Hulshoff Pol, 2010; Rosazza
and Minati, 2011; Smith et al., 2009; Smith et al., 2013; Spreng et al., 2012].
Connectivity in rsfMRI can be divided into functional connectivity and effective connectivity
[Friston, 1994; Friston, 2011]. This literature review focuses on functional connectivity due to the
paucity of research using effective connectivity to look at clinical populations to date. Functional
connectivity describes the temporal correlation between spatially segregated events and it does not
contain information about directionality, meaning that the analysis reports only the covariance
between brain regions/voxels (i.e. how much the neural activity changes in different locations are
related to each other). Several approaches have been used to assess rsFC in healthy subjects and
patients with neurological conditions to explore brain organization and neuroimaging biomarkers,
respectively [Bowman, 2014b; Lindquist, 2008; Margulies et al., 2010; Pievani et al., 2014].
9
1.2.1.1 Common analyses and applications
Several common approaches have been applied to clinical studies such as Independent Component
Analysis (ICA), seed-based correlation, whole brain correlation, and graphical analysis. In brief,
ICA decomposes the data into maximally spatially independent components (spatial maps) and
corresponding time courses [McKeown et al., 2003]. Voxels within the same maps may be
considered functionally connected. Maps may correspond with areas associated with the default
mode network, sensorimotor processing, visual processing and saliency [Damoiseaux et al., 2006].
Another approach for determining rsFC is to utilize a seed-based approach. In this technique, one
must pre-specify seed region(s) and correlation between the time course of the seed region(s) and
other voxels are calculated [Greicius et al., 2003]. Whole-brain correlational analysis takes into
account several brain regions at the same time and evaluates the covariance/correlation between
regions rather than looking at specific seed regions [Prodoehl et al., 2014].
Once connectivity strengths between brain regions have been computed, the results are usually
collected into a matrix and summarized. One such summarizing method is graph theory. In
mathematics, a “graph” consists of nodes and edges, and in this context, the nodes represent brain
regions and the edges represent the statistical relation between their respective time courses (e.g.
correlation, possibly after thresholding and setting small correlational values to zero). Recently,
graphical analyses have become a popular tool to describe how networks are organized and how
the information is coordinated through biological fundamentals in connectome studies [Rubinov
and Sporns, 2010]. Topological features reveal that the brain is a small-world organization (i.e.
networks are highly clustered like a scale-free network but nodes are linked by small path lengths
like a random network) with a balance between how well the information is processed within one
10
system (i.e. functional segregation) and how efficient the information is integrated throughout the
brain (i.e. functional integration) [Deco et al., 2015; Sporns, 2013]. These organization principles
aim to reduce wiring cost yet enhance the efficiency of information flow [Bullmore and Sporns,
2012; Sporns, 2013]. As the volume of the human brain has increased during evolution, neural
signaling is required to travel longer distance along white matter, resulting in longer travel time
(i.e. wiring cost) [Hofman, 2014]. Other adaptations have also been proposed as strategies to
reduce wiring cost in addition to segregation, integration, and small world organization. As the
travel distance increases, “short cuts” of the brain have been developed such as long-range
connections so neuronal signals can bypass other regions through these short cuts [Hofman, 2014].
In addition, studies have proposed that several regions are more densely connected to other regions
either within one network or between networks, forming high traffic spots in the brain called
“hubs” [Achard et al., 2006; van den Heuvel and Sporns, 2013; Power et al., 2013]. Therefore,
neuronal signals can “transfer” between these hubs to reduce wiring cost as the information flow
can be more efficiently carried out across the brain.
1.2.2 Dynamic functional connectivity
The above-mentioned methods assume that connectivity patterns are relatively temporally
stationary and estimate the average connectivity patterns that represent neurophysiological events
across the scanning time. However, new studies with computational models implemented have
suggested that functional connectivity is not stationary and this aspect has been referred to as
dynamic functional connectivity [Hutchison et al., 2013b; Hutchison et al., 2013a], i.e., the
neuronal patterns fluctuate across time (in seconds/minutes) in order to maintain brain function in
11
response to different environmental stimuli. This temporal connectivity pattern is also a crucial
component of brain organization.
1.2.2.1 Methods and applications
The most common analytical model for dynamic functional connectivity is perhaps a sliding
window approach, with pairwise correlation implemented [Hutchison et al., 2013b; Hutchison et
al., 2013a]. This model estimates correlations between brain regions within a fixed-length, sliding
window, with the (possibly overlapping) windows ultimately moved over the entire data. As a
result, each correlation matrix represents the connectivity at each window. With post hoc analyses,
features that summarizes connectivity changes across time can be calculated [Liao et al., 2015].
Another approach is to apply clustering methods (i.e. to separate matrices into different states) and
calculate the “dwell time” of each state [Damaraju et al., 2014]. Alternatively, principal component
analysis (PCA) can be applied to study whole brain dynamics based on windowed correlation
matrices [Leonardi et al., 2013]. Nevertheless, there are potential pitfalls with sliding window
approaches [Hindriks et al., 2016; Hutchison et al., 2013b]; if the window is too long, important
dynamic changes may be missed. If the window is too short, the connectivity estimates may be
unstable as too few samples are available for the statistical inferences. A window length of 30-60
seconds for fMRI data has been heuristically suggested [Leonardi and Van De Ville, 2015; Zalesky
and Breakspear, 2015]. Other time-varying approaches have also been implemented to investigate
dynamic functional connectivity such as the combination of sliding window and Hidden Markov
Model [Chiang et al., 2016], the Sticky Weighted Regression Model [Liu et al., 2015], and
Dynamic Conditional Correlations [Lindquist et al., 2014]. Other methods which investigate
dynamics in frequency domain and integrate with graph analysis have been proposed [Chang and
12
Glover, 2010; Chiang et al., 2016]. A comprehensive review has pinpointed the strengthens and
issues of each method [Preti et al., 2017]. Although some of them have been applied to clinical
studies, the majority studies which explore the links between disease, cognition, and dynamic
functional connectivity still favour the sliding window approach.
Dynamic functional connectivity allows for different degrees of engagement in areas/networks
across different time scales (i.e. a given brain area gets involved in the connectivity network at a
certain time point, but in a different time frame the area is not engaged in the network). Studies in
behaviour and simultaneous recording of functional connectivity and electrophysiological data
suggest that temporal variations link neuronal origin, cognitive processes and behaviour [Chang
et al., 2013; Thompson et al., 2013]. For example, increased temporal variability of neural activity
is associated with stable performance in an electroencephalography (EEG) memory task [McIntosh
et al., 2008]. Moreover, recently both rsfMRI and tasked fMRI research has further suggested that
these dynamic and topological features have been linked to cognitive architecture -- especially
higher-order cognitive processes [Mattar et al., 2015; Shafto and Tyler, 2014]. Compared to motor
tasks, the performance of the N-back memory task requires more functional integration between
regions and relies on the ability to flexibly integrate information across regions and networks
especially in frontoparietal and frontotemporal networks [Braun et al., 2015; Shine et al., 2016].
Such brain organization has been seen in rsfMRI as well, whereby the rsFC network related to
cognition requires flexibility [Mattar et al., 2015]. In addition, it has been concluded that dynamic
interaction between networks supports cognitive functions, especially ones which requires
complex processes such as language [Chai et al., 2016; Shafto and Tyler, 2014].
13
1.3 Resting-state functional connectivity and executive function in Multiple Sclerosis and
Parkinson’s Disease
Table 1.1 summarizes the studies which have explored the associations between executive deficits
and rsFC in MS and PD. Figure 1.1 visualizes the rsFC patterns related to executive function.
1.3.1 Multiple Sclerosis
Many studies have reported decreased functional connectivity in MS in resting-state networks
(RSNs) including the default mode, salience, executive control, working memory, and
sensorimotor networks [Bonavita et al., 2011; Rocca et al., 2012]. The decreased connectivity
indicates that cortical regions fail to integrate information (especially the medial prefrontal cortex
and posterior cingulate cortex) and has been associated with worsening executive function in
cognitively impaired patients [Cruz-Gómez et al., 2014; Louapre et al., 2014], which supports the
idea that cognitive deficits in MS may be the result of disconnection. In addition, studies which
investigate whole-brain rsFC have revealed weaker connectivity among widespread regions
including frontal, parietal, temporal, and subcortical areas as well as interhemispheric connectivity
[Richiardi et al., 2012; Zhou et al., 2013].
Research with novel approaches such as graph theoretical analysis provides insight into alterations
of functional connectivity in a topological fashion [Gamboa et al., 2014; Rocca et al., 2016b;
Schoonheim et al., 2013]. Patients with MS exhibit reduced global efficiency, node degree,
centrality, and increased path length on average compared to normal subjects, demonstrating
altered rsFC in network organization [Rocca et al., 2016b]. As global efficiency calculates the
average inverse of shortest path lengths, the reduced measure indicates that the shortest path length
14
increases, and thus a marker of inefficient information transfer in the brain. Decreased node degree
suggests that brain regions lose connections to/from other regions, indicating a reduction in overall
connectivity. Centrality represents the “importance” of a given region (i.e. the fraction of shortest
paths that pass through a given node) and a small value implies that 1) the region loses shortest
paths and 2) hubs potentially vanish. As a consequence, the organized brain network structure,
which is meant to reduce wiring cost, fails. These topological changes in MS illustrate impaired
functional integration and the impairment is associated with executive dysfunction such as reduced
accuracy in dual-task performance [Gamboa et al., 2014; Schoonheim et al., 2013]. This implies
that the MS brain becomes more segregated and will be less efficient to support complex
behaviour. In addition, the loss of hub regions in MS (include the superior frontal gyrus, precuneus,
and anterior cingulate cortex) leads to impaired functional integration, as hubs are an important
aspect of functional integration where several networks are integrated [Rocca et al., 2016b]. As
many of these regions are highly related to executive function, the results imply that higher
cognitive function requires not only the frontal cortex but other cortices, such as the parietal cortex
in case of the precuneus, to coordinate information flow.
However, other studies have also reported that increased functional connectivity in some RSNs is
associated with reduced cognitive efficiency, indicating that cognitive deficits may be associated
with enhanced functional coupling in MS possibly reflecting compensatory and/or adaptive
mechanisms at different stages of disease progression [Faivre et al., 2012; Hawellek et al., 2011b].
Moreover, the authors reported that the major connectivity which modulated behaviour required
shifts between the default mode network and the central executive network depending on the
degree of patient’s cognitive impairments, in which connectivity shifted toward executive network
15
in patients with higher cognitive efficiency [Hawellek et al., 2011b]. The networks with increased
connectivity lost “flexibility” in network interactions and resulted in stronger connectivity at a
longer time scale. As these networks commonly require coordinated information flow across
cortices via long-range fibers, it has been postulated that cognition-related connections may
constantly shift between regions and increased functional connectivity reflects a loss of diversity
in network interactions.
With a seed-based approach, other studies have demonstrated that increased functional
connectivity, which was observed in the frontal regions, anterior cingulate cortex, and posterior
areas, was related to better performance of executive ability [Loitfelder et al., 2012]. As the seed
regions in the studies are all important hubs in cognitive processes (i.e. the anterior cingulate
cortex, ventral medial prefrontal cortex, and posterior cingulate cortex), the results imply that core
cognitive regions may first demonstrate compensatory effects. Yet these studies focus on a limited
number of seed regions, which may lose sensitivity to detect connectivity patterns in other regions.
In addition, studies often report both increased and decreased activation as a function of disease
stage and cognitive impairments. Perhaps rather than increased or decreased activity per se, the
“interactions” between networks with altered activation/connectivity would be more informative
to cognitive impairments and disease severity. Therefore, studying whole brain connectivity may
prove beneficial. Effects of age and disease stage on cognitive deficits may also lead to inconsistent
results. This may be related to cognitive reserve capacity whereby in early stages of the disease
there is altered connectivity and normal cognitive performance, but over time or under higher
cognitive demands, the capacity for compensation diminishes and the altered connectivity
becomes aberrant.
16
1.3.2 Parkinson’s Disease
The majority of research into rsFC in PD focuses on the connectivity between the motor cortex
and subcortical areas such as cortico-striatal connections as well as how rsFC correlates with
clinical scores [Helmich et al., 2010; Kwak et al., 2012; Luo et al., 2014; Sharman et al., 2013;
Wu et al., 2009]. Since non-motor symptoms such as cognitive deficits in PD have profound effects
on quality of life, a few studies have started investigating whole-brain functional connectivity and
its association with cognitive deficits [Amboni et al., 2014a; Baggio et al., 2014; Disbrow et al.,
2014; Hirano et al., 2012; Olde et al., 2014; Owen, 2004].
Through a seed-based approach with a seed in the caudate, demented PD patients exhibited
decreased connectivity to frontal regions [Seibert et al., 2012], suggesting that alterations in the
connections between frontal and subcortical regions are important contributors to dementia in PD.
Research on whole brain functional connectivity showed that the inferior frontal gyrus and
posterior regions of the brain (e.g. superior parietal lobes and multiple occipital regions) had
decreased connectivity and demonstrated stronger correlation with global cognitive function
(including orientation, language, memory, praxis, attention, abstract thinking, perception and
calculation) than motor deficits [Olde et al., 2014]. Connectivity alterations in frontal and parietal
areas have also been shown to be associated with mild cognitive impairment (MCI) in PD [Amboni
et al., 2014b]. Furthermore, decreased activity in the ICA-derived default mode network was
independent of patients’ cognitive states but decreased frontoparietal connectivity was associated
with worsening visuospatial, memory, and executive functions [Amboni et al., 2014b]. However,
another study has proposed the opposite pattern: namely that executive performance in PD is
related to the default mode network rather than the executive control network [Disbrow et al.,
17
2014]. In addition, inter-network connectivity (i.e. correlations between RSNs) between the default
mode network (DMN), salience network (SN), and central executive network (CEN) was altered
in PD with increased coupling between the DMN/CEN and decreased coupling between the
CEN/SN in PD with a particularly reduced interaction between the SN and striatum [Putcha et al.,
2015].
The discrepancies between rsFC in PD imply that a more robust measure across cohorts is needed
to study how connectivity is related to cognition. Graph-theory based analysis of rsfMRI data
suggests that PD manifests as a disconnection syndrome [Göttlich et al., 2013]. Graphical
measures demonstrated that PD showed lower connectivity (lower node degree) in the medial and
middle orbitofrontal cortex, while the superior parietal, posterior cingulate, supramarginal, and
supplementary motor areas presented increased connectivity (higher node degree), the latter being
interpreted as compensation for cortico-striatal dysfunction. The connections between subcortical
areas and orbitofrontal/temporal regions as well as within cortical regions were weaker in the PD
group, indicating that PD is associated with widely disrupted resting-state networks [Göttlich et
al., 2013]. Baggio et al. investigated the relationship between graphical measures and cognition in
PD [Baggio et al., 2014]: MCI-PD subjects presented with increased modularity and clustering
coefficients. Modularity quantifies the degree to which the network may be subdivided into
separate groups and it is an index of functional segregation, i.e. how well the information is carried
out within one system. Clustering coefficient measures the degree to which nodes in a network
tend to gather together, which is also a measure of functional segregation. In PD, functional
connectivity has become more segregated as the long-range connections were affected between
frontal, parietal, and occipital areas. Hub reorganization has been identified in MCI-PD, where
18
normal hubs in the parietal and superior temporal lobes lose their central role and non-hubs in the
middle frontal gyrus gain importance, implying an association between cognitive deficits and hub
distribution [Baggio et al., 2014]. Furthermore, these graphical measures were correlated with test
scores of worsening attention, executive function, memory, and visuospatial functions in MCI-PD
compared to control subjects, indicating that segregated brain networks measured by graphical
approaches reflect decreased cognitive ability including executive functions. Hub reorganization
has been described in another study where patients with PD no longer had prominent connectivity
in the insula but caudate connectivity became more prominent, however, no cognitive scores were
related to the changes [Koshimori et al., 2016].
To summarize, weakened connections between frontal and other regions are associated with
cognitive dysfunction in PD. Although many studies have shown that cognitive deficits are related
to widespread dysconnectivity between frontal and other cortical regions (i.e. posterior and
occipital areas), we cannot exclude the importance of frontostriatal circuits as the impaired
dopamine pathways may cause inefficient feedback to frontal areas, leading to cognitive
dysfunction [Owen, 2004]. In fact, studies have reported the links between impaired dopamine
signaling in the frontal areas and executive dysfunction such as working memory and set-shifting
deficits [Narayanan et al., 2013]. In addition to dopaminergic system, cholinergic dysfunction
(which affects frontal, temporal, occipital, and forebrain areas) has been suggested as another
contribution to executive dysfunction [Narayanan et al., 2013; Pagonabarraga and Kulisevsky,
2012]; therefore, investigating distributed regions is necessary rather than focusing on specific
circuits. Meanwhile, the finding of increased connectivity patterns, which have been interpreted
as compensatory effects, is not consistent across studies. We speculate that this might be because
19
the nature of dynamic functional connectivity in PD is not captured by traditional analyses and the
appropriate emphasis of “hub regions” has been neglected. As switching ability is impaired in PD
[Cools et al., 2001], patients demonstrate cognitive inflexibility and perhaps neuronal inflexibility,
which is implied by the studies showing that dopamine signaling can impact dynamic neural phase
coordination and play a role in cognitive processes [Dzirasa et al., 2009]. Therefore, we
hypothesize that how rsFC changes (i.e. network dynamics or dynamic functional
connectivity) together with how hub regions and networks are remapped might be a feature
which reflects cognitive rigidity in PD. Further studies with time-varying approaches and
graphical analyses are necessary to reveal hidden patterns in rsFC.
20
Disease Authors Methods for
rsFC
Purposes and
methods for
neuropsychological
tests
Subjects Main findings in patients
Multiple
Sclerosis (MS)
Bonavita et al.,
2011
ICA-RSNs determine CI and
CP
CI-RRMS, CP-
RRMS, NC
DMN showed decreased rsFC in the anterior
cingulate cortex but increased rsFC in the posterior
cingulate cortex
Rocca et al., 2012 ICA-RSNs NA RRMS - decreased rsFC in the salience, executive control,
working memory, sensorimotor, visual networks as
well as DMN
- increased rsFC in the executive control and
auditory networks
Cruz-Gómez et al.,
2014
ICA-RSNs - determine CI and
CP
- regression analysis
CI-RRMS, CP-
RRMS, NC
- All RSNs in CI decreased compare to NC and CP
- CP showed decreased rsFC in the left
frontoparietal network
- rsFC was related to radiological variables not
cognition
Louapre et al.,
2014
ICA-RSNs determine CI and
CP
CI-RRMS, CP-
RRMS, NC
- increased rsFC in CP at the attention network
- decreased rsFC in CI at DMN and the attention
network especially between the medial PFC and
PCC
Richiardi et al.,
2012
Pearson’s r
whole brain
matrix and
classification
NA RRMS, NC the connections in subcortical and fronto-parieto-
temporal areas were the most discriminative as well
as inter-hemispheric connections
Zhou et al., 2013 Pearson’s r in
symmetric
interhemispheric
voxels
NA RRMS, NC - decreased rsFC in high-order cognitive regions
including frontal, temporal, and occipital regions
- increased rsFC in the OFC, insula, thalamus,
cerebellum, subcortical, and inferior temporal areas
Gamboa et al.,
2014
graph theory –
the Brain
Connectivity
Toolbox
correlation analysis RRMS, CIS, NC - reshaped modules in MS were negatively
correlated with the accuracy of dual task
performance and indicated impaired functional
integration
Rocca et al., 2014 graph theory -
the Brain
Connectivity
Toolbox
determine CI and
CP
RRMS, benign
MS, SPMS, NC
- MS lost hubs in the superior frontal gyrus,
precuneus, and anterior cingulum but gain hubs in
the temporal pole and cerebellum
- decreased degree, clustering coefficient, global
efficiency, hierarchy as well as increased path
21
length and assortativity in MS indicated impaired
functional integration
- functional integration was more impaired in CI
Schoonheim et al.,
2013
graph theory –
eigenvector
centrality
mapping
correlation analysis RRMS, PPMS,
SPMS, NC
- increased centrality/importance in the thalamus
and PCC
- decreased centrality/importance in sensorimotor
(related to EDSS) and ventral stream areas (related
to poor executive function and processing speed)
Faivre et al., 2012 ICA-RSNs correlation analysis RRMS, NC - increased rsFC in DMN and the fronto-parietal
network was correlated with decreased semantic
fluency and increased PASAT scores
Hawellek et al.,
2011
ICA-RSNs factor analysis and
correlation analysis
relapse-free MS,
NC
higher rsFC in the cognitive control network and
DMN was related to reduced executive function
(set-shifting)
Loitfelder et al.,
2012
seed-based multiple regression
model
CIS, RRMS,
SPMS, NC
- increased rsFC in the ACC, PCC, angular gyrus,
and postcentral gyrus
- better executive abilities were related to increased
rsFC in the cerebellum, middle temporal gyrus,
occipital areas, and angular gyrus
Wojtowicz et al.,
2014
seed-based act as covariates in
connectivity
analysis
RRMS, NC better processing speed was related to greater rsFC
between the ventral PFC and left frontal pole
Parkinson’s
Disease (PD)
Helmich et al.,
2010
seed-based NA PD, NC - decreased rsFC between the posterior putamen and
motor areas/inferior parietal cortex
- increased rsFC between the anterior putamen and
inferior parietal/temporal cortex
(this study was not related to executive function)
Kwak et al., 2012 amplitude of
low frequency
fluctuations
(*measure fMRI
signal intensity)
measure overall
cognition
PD, NC - with OFF medication, increased rsFC was seen in
the primary and secondary motor areas,
middle/medial prefrontal cortex
- with ON medication, reduced rsFC in the
prefrontal and motor cortical areas
- poor motor task performance was associated with
higher rsFC
(this study was not related to executive function)
Luo et al., 2014 seed-based measure overall
cognition
PD, NC - reduced rsFC in the mesolimbic-striatal and
corticostriatal loops
22
- decreased functional integration across striatum,
mesolimbic cortex, and sensorimotor areas
- higher non-motor symptom indices were related to
reduced rsFC in amygdala-putamen circuits (but not
restricted to executive function)
Sharman et al.,
2013
seed-based NA PD, NC - increased rsFC in the limbic and associative
cortex, limbic cortex and thalamus, putamen and
thalamus
- decreased rsFC in the sensorimotor cortex and
thalamus, globus pallidus and putamen/thalamus,
substantia nigra and globus pallidus/putamen/
thalamus
(this study was not related to executive function)
Wu et al., 2009 regional
homogeneity
measure overall
cognition
PD, NC - decreased rsFC in the putamen, thalamus, and
supplementary motor areas
- increased rsFC in the cerebellum, primiary
sensorimotor cortex, and premotor area
(this study was not related to executive function)
Seibert et al., 2012 seed-based NA PDD, PD, NC - PDD showed decreased rsFC in middle frontal
regions
Olde et al., 2014 synchronization
likelihood for
whole brain
connectivity
generalized
estimated equations
(combine correlation
and linear
regression)
PD, NC - decreased rsFC in the inferior frontal gyrus,
superior parietal lobs, and occipital regions showed
stronger correlations to cognitive function including
executive function rather than motor symptoms
Amboni et al., 2014 ICA - RSNs correlation analysis PD, MCIPD, NC - decreased DMN in PD and MCIPD was not
correlated with cognition
- decreased rsFC in the frontoparietal network in
MCIPD was related to worse executive function but
not clinical variables
Disbrow et al.,
2014
seed-based - measure cognitive
profile
- correlation analysis
PD, NC - No changes in the executive control network
- decreased rsFC within DMN was seen in PD and
greater DMN rsFC was related to faster processing
speed
Putcha et al., 2015 ICA – RSNs and
Pearson’s r
measure overall
cognition
PD, NC - interactions between RSNs were altered in PD
- decreased interactions between the salience and
executive networks
23
- increased interactions between the DMN and
executive network
Göttlich et al., 2013 graph theory –
the Brain
Connectivity
Toolbox
NA PD, NC - lower node degree in the medial/middle OFC and
occipital pole
- increased node degree in the superior parietal,
PCC, supramarginal, and supplementary motor
areas
- decreased rsFC between the subcortical, medical
OFC and temporal regions
Baggio et al., 2014 graph theory -
the Brain
Connectivity
Toolbox
- determine PD and
MCIPD
- linear regression
analysis
MCIPD, PD, NC - both MCIPD and PD showed reduced rsFC in
long-range connections which include frontal,
occipital, and parietal regions
- increased local rsFC in MCIPD including higher
clustering, local efficiency, and modularity in the
frontal areas was correlated attention/executive
scores
- MCIPD showed reduced importance in hubs
Koshimori et al.,
2016
seed-based and
graph theory –
Graph-
Theoretical
Analysis
Toolbox
correlation analysis PC, NC - hub reorganization was seen in PD, which PD lost
the insula but gained the caudate as a new hub in
cognitive networks
- increased rsFC in the DLPFC and insula of
cognitive networks
- reduced intrahemispheric rsFC within insula and
supplementary motor areas was related to
medication not cognition
Table 1.1 Summary of the rsFC studies in MS and PD. The inclusion criteria for the studies are 1) performing neuropsychological tests, 2) using
statistical analyses to link fMRI measures and cognitive scores and 3) the reported results are associated with executive performance or dementia if the
neuropsychological tests are used for determining dementia.
[NC: normal controls, CI: cognitive impaired, CP: cognitive preserved, DMN: the default mode network, ICA – RSNs: independent component analysis
derived resting state networks, RRMS: relapsing-remitting MS, SPMS: secondary progressive MS, CIS: clinically isolated syndrome, PASAT: Paced
Auditory Serial Addition Test, PDD: demented PD, MCIPD: mild cognitive impairment PD, PFC: prefrontal cortex, PCC: posterior cingulate cortex,
ACC: anterior cingulate cortex, OFC: orbitofrontal cortex, DLPFC: dorsal lateral prefrontal cortex]
24
Figure 1.1 The connections and regions of rsFC correlated with executive performance in healthy subjects
(HS), multiple sclerosis (MS), and Parkinson’s disease (PD). The left panel shows the rsFC in HS from
[Hampson et al., 2006; Reineberg et al., 2015; Reineberg and Banich, 2016; Seeley et al., 2007]. The upper-right
panel shows the rsFC in MS with a small brain indicating the pathological features (lesions in white matter).
The lower-right panel demonstrates the rsFC in PD and the small brain indicates the pathology (affected
dopamine pathways). In MS and PD, node size explains whether the region obtains or loses hub identity.
Small nodes indicate a loose of hub identity, while big nodes indicate a gain of hub identity. The nodes with
medium size represent unchanged hub identity. The maps are generated with the BrainNet Viewer
(http://www.nitrc.org/projects/bnv/) based on the studies in Table 1.1.
25
1.3.3 Summary
It is becoming increasingly common to interpret cognitive performance as a result of coordination
of information between brain regions and rapid re-configuration among connectivity states
[McIntosh, 2000; Shaw et al., 2015; Sridharan et al., 2008]. Regardless of distinct pathology, early
rsFC studies either applied exploratory methods to discover specific brain areas and networks, or
they focused on pathways based on prior hypotheses related to the disorders, which has led to a
paucity of research exploring the role of other regions and networks. For example, the default
mode network has been intensively studied in MS, and most studies have reported decreased
connectivity. Functional connectivity within frontostriatal loops in PD has been extensively
examined due to affected dopaminergic pathways and reduced connectivity in the loops has been
linked to clinical deficits. However, with the approaches which assess whole brain functional
connectivity, widespread connectivity alterations have been revealed rather than abnormalities in
just one circuit.
1.4 Impacts and contributions of studies on resting-state functional connectivity to
clinical neuroscience
1.4.1 Key connections in cognitive deficits
Long-range connections, such as frontal and parietal links, seem to play an important role in
executive functions regardless of distinct neurological impairments. From an evolutionary
perspective, the human brain has increased in size dramatically and intrahemispheric connections
have been shown to be stronger in larger brains [Hänggi et al., 2014; Hofman, 2014]. In addition,
it has been proposed that long range communication networks in humans linking prefrontal areas
26
to other regions, such as parietal areas, are important adaptations in evolution compared to other
species, and these connections are necessary for complex behaviour and sophisticated cognitive
abilities in humans [Hofman, 2014; Shaw et al., 2015].
A review of neurological disorders based on network analyses has proposed that hub overload and
failure potentially affect long-range connections by disrupting the normal hierarchical architecture
of connectivity networks (i.e. imbalance between local process and global process via hubs) and
the changes are related to executive dysfunction, which may be a common pathway leading to
cognitive impairments in neurological disorders [Stam, 2014]. However, how hub reorganization
alters spatial and temporal coordination between brain regions through long-range connections
(i.e. flexibility in long-range connections) remains unclear. Long-range connections propagate
neural information, which is essential for high-order cognition between segregated brain regions,
and these information flows can be better organized and coordinated through hubs so that wiring
cost is minimized. In order to investigate these associations, more research on the links between
brain topology and dynamic functional connectivity is required.
The idea proposed by Stam [Stam, 2014] is consistent with the studies we have discussed in the
review sections of this chapter, in which many investigations have reported impaired long-range
connectivity between frontal, parietal, and occipital regions in different neurological conditions.
Research has also demonstrated the associations between frontal-parietal connectivity and high-
order cognitive processes including executive function, emphasizing the importance of long-range
connections in human cognition. For instance, a study which combined fMRI meta-analysis and
DTI tractography revealed that a ventral branch of the superior longitudinal fasciculus, which
connects inferior frontal and inferior parietal lobes, is associated with several cognitive functions
27
such as working memory, decision making, language process, inhibition, and attention [Parlatini
et al., 2016]. Moreover, another study has demonstrated that diffusion measures in the superior
longitudinal fasciculus are correlated with performance in executive function, information
processing speed, and memory domains [Sasson et al., 2013]. These studies further emphasize the
strong links between long-range connections and cognition.
Long-range connections are impaired in neurological disorders, and may be inferred by structural
MRI studies. Cortical thinning in frontal and parietal cortices has been correlated with worse
executive function in PD, while poor executive performance can be predicted by cortical thinning
in the anterior cingulate cortex in MS [Duncan et al., 2016; Geisseler et al., 2016]. In addition,
based on our unpublished studies, higher myelin integrity in the bilateral inferior fronto-occipital
fasciculus, inferior longitudinal fasciculus, and superior longitudinal fasciculus, which connect
frontal/parietal/occipital areas, is associated with better executive function in MS, supporting that
long-range connections are important for higher-order functions [Baumeister et al.,]. Therefore,
the literature suggests that executive dysfunction may be partially caused by disruptions between
frontal regions and the posterior part of the brain (especially frontal and parietal cortices) in both
neurological conditions presented in this chapter.
1.4.2 Three components of executive function
Why would the disconnection between frontal and parietal regions contribute to executive
dysfunction in neurological conditions? In other words, what are the roles of frontal and parietal
regions in executive function? Research has concluded that the prefrontal cortex (PFC) is a core
area in executive function, but it also relies on the inputs from other regions such as the posterior
cortex to facilitate the process [Bunge and Souza, 2009; Carpenter et al., 2000; Miller and Cohen,
28
2001; Miller and Wallis, 2009]. Miller and Cohen have proposed a model of executive function,
which includes three components of neuronal activity: inputs, internal states, and outputs [Miller
and Cohen, 2001]. The PFC receives sensory inputs such as visual, somatosensory, and auditory
information from the occipital, temporal, and parietal regions as the PFC is connected to secondary
or association sensory cortex. The intrinsic connections communicating between sub-regions in
the PFC allow information to be processed and distributed to other sub-regions, which facilitates
the core process (i.e. intensive “thinking”) of executive function. Finally, the processed
information has to be executed through motor outputs such as motor association areas and
subcortical regions as recent research states [Monchi et al., 2006; Owen, 2004]. Therefore, the
connections between frontal regions and sensory cortices become indispensable as these are the
inputs for executive processes. Of note, the central executive network identified by rsfMRI
includes both prefrontal areas as well as parietal areas. Only a few of the studies included in this
chapter report connectivity alterations in the temporal cortex [Richiardi et al., 2012], and there is
not enough information to conclude whether connectivity in the temporal areas is strongly
associated with executive functions in PD and MS. Although several imaging studies have reported
cortical changes in the temporal areas, not all the cortical alterations show strong correlation to
executive dysfunction [Achiron et al., 2013; Möller et al., 2016; Tam et al., 2005]. Overall, the
results of connectivity in temporal regions are inconsistent. Therefore, we concluded that impaired
long-range connections bridging frontal, parietal, and occipital areas possibly damage the "inputs”
component, resulting in executive dysfunction seen in clinical presentation (Figure 1.2).
29
Figure 1.2 A model of executive function. This diagram incorporates anatomical information into the model
proposed by Miller and Cohen [Miller and Cohen, 2001] and illustrates how information flows between the
three components: inputs, internal states, and outputs. The blue curved arrows represent the sensory inputs
coming from parietal, occipital, and temporal lobes to the prefrontal cortex. The yellow gradient illustrates
where the core process happens in internal states, which is the prefrontal cortex. Red straight arrows show
the outputs are sent to subcortical areas to execute (thick red arrow) as well as the feedback information
(thin red arrow) coming from subcortical regions, which are labeled with red gradients. The thin red double
arrow indicates the coordination between subcortical regions and the motor association cortex.
30
Another well-known model of executive function is the dorsolateral prefrontal circuit, which is
one of the frontal-subcortical circuits [Bonelli and Cummings, 2007]. The dorsolateral prefrontal
circuit mediates executive function through the neurons which originate in the prefrontal lobes and
project between the dorsolateral prefrontal cortex, caudate nucleus, global pallidus, and thalamus.
Early evidence that patients with lesions in this loop demonstrated executive deficits measured in
neuropsychological assessments emphasized the role of the dorsolateral prefrontal circuit in
executive function [Bonelli and Cummings, 2007]; however, the model neglects regions which are
not directly linked to this circuit, potentially providing an over-simplified viewpoint of executive
functioning in humans.
Working memory has been proved to be strongly correlated with executive function and also relies
on connections in prefrontal-parietal regions. For example, in neuroimaging studies, the
dorsolateral prefrontal cortex and parietal regions were involved in both working memory and
executive tasks [Carpenter et al., 2000], while behavioural studies showed the correlation between
working memory capacity and executive function was higher than that between other domains
[McCabe et al., 2010]. However, executive function is not the same as working memory even
though certain activation patterns overlap. In fact, many tasks which assess executive function also
require working memory ability, which makes it difficult to isolate working memory from
executive function explicitly. The literature presented in this chapter suggests that long-range
connections between prefrontal and parietal regions are essential for executive function in
neurological disease, and are also important for working memory. Further neuroimaging research
is needed to profile the neural patterns of these domains, such as network dynamics and cortical
coordination.
31
1.4.3 Unique features of brain organization in different neurological conditions may all
result in executive dysfunction
We have proposed that impaired long-range connectivity might be one of the common causes of
executive dysfunction in MS and PD. In addition to this common impairment, there are unique
features which may contribute to cognitive deficits. As executive function requires flexibility to
switch approaches [Diamond, 2013], it has been shown that the neuronal connectivity of executive
function involves abilities to dynamically recruit different brain regions and coordinate
information in a flexible manner, including or excluding networks based on the neuronal need to
overcome the trade-offs between wiring cost and connectivity [Braun et al., 2015; Bullmore and
Sporns, 2012; Mattar et al., 2015]. This brain organization principle, which has been mentioned
previously (section 1.2.2) as dynamic functional connectivity, has been referred to as one of the
fundamental aspects of higher-order cognition (i.e. the cognitive processes that require intensive
thinking and planning) [Bressler and Scott Kelso, 2001; Hutchison et al., 2013a; Stephens et al.,
2013]. Furthermore, as human brain volume has increased in evolution, the efficiency of
information transfer in long-range connections has become more important for cognition [Hofman,
2014]. Proposals have been made that hub regions are formed to better propagate information flow
between distinct brain regions by reducing wiring cost and these hubs are especially active in
higher-order cognition [Baggio et al., 2015; van den Heuvel and Sporns, 2013; Hofman, 2014;
Power et al., 2013]. Altered hub organization reduces efficiency of information flow between
anatomically segregated regions in MS, contributing to cognitive deficits. The executive deficits
in PD are perhaps also caused by impaired frontostriatal connectivity [Owen, 2004], which may
damage the “output” component of execution function. Moreover, abnormal brain organization
32
such as decreased dynamic connectivity and hub changes may contribute to executive dysfunction
as well.
Taken together, both hub organization and dynamic functional connectivity are key factors to
facilitate higher-order cognition including executive function. Losing dynamic coordination and/or
efficiency of information transfer may contribute to clinical deficits including cognitive
dysfunction in several brain disorders [Fornito et al., 2015]. Therefore, we argue that dynamic
functional connectivity, impaired in PD due to neurotransmitter dysregulation, probably causes
executive dysfunction; while altered hub organization in these two conditions contribute to
executive deficits unequally as the affected hubs are not exactly the same.
1.5 Research objectives and thesis organization
Traditional research which investigates functional connectivity and cognitive decline in PD and
MS has focused on specific loops and regions. The approaches that have thus far been applied
neglect that 1) compensation effects may occur in many locations of the brain, 2) higher-order
cognitive function, which appears to impact quality of life the most, requires not only frontostriatal
loops but other remote regions, and 3) temporal variation of functional connectivity potentially
plays a prominent role in cognition. Several studies utilizing advanced network analyses to
investigate brain organization in healthy and disease populations have demonstrated promising
preliminary results. These analyses on whole brain rsFC have revealed that functional integration,
functional segregation, dynamic functional connectivity, and hub structures are necessary and
indispensable to executive functioning. However, clinical research that applies these analyses to
study the associations between executive function, clinicopathology, and rsFC remains limited. In
this dissertation, the goal is to highlight the importance of brain organization to executive
33
function with the use of advanced network analyses (e.g. graph theoretical theory and time-
varying approaches) and explore the relations between cognition, clinicopathology, and
brain organization in two representative neurological conditions. Figure 1.3 illustrates the
overall flow of the thesis.
Figure 1.3 A diagram of the flow in this thesis research. Red and blue areas represent
cognitive impairments in MS and PD, respectively. The overlapping area in purple represents
the common cognitive deficits: executive dysfunction.
In the following chapters, we utilize different data sets:
34
- The Parkinson’s Progression Marker Initiative (PPMI). This is a publically-available data set
funded by The Michael J. Fox Foundation for Parkinson’s Research, containing advanced imaging,
biologic sampling, clinical and behavioural assessments with the goal of identifying biomarkers
of Parkinson’s disease progression. We included one cohort in the PPMI database which contained
31 subjects as all the subjects were scanned in the same medical centre. This data set is included
in chapter 2 and 4.
- The COGMS data set. This data set was generated from a MS-society grant (awarded to
MacKay and McKeown). The original design was to evaluate advanced structural imaging
features, resting-state fMRI measures, and cognitive profile in subjects with MS and combine three
components together to study the relations between structural integrity, functional connectivity,
and cognitive performance (particularly executive function). Forty-six MS subjects were included
in chapter 2,4,5, and 6 as well as a subset of age-matched MS and control subjects.
- The BCT data set. This was data acquired in Pacific Parkinson’s Research Centre (PPRC) and
the original design included task-driven fMRI and resting-state fMRI to evaluate the functional
connectivity under a motor task and at rest rather than brain-cognitive relations. Therefore, there
were limited cognitive measures. Twelve PD subjects and ten healthy controls were included and
the data was used in chapter 3.
- The OPERA data set. This was a phase III clinical trial conducted in MS clinic at UBC Hospital
and supported by F. Hoffmann-La Roche, Ltd. The purpose was to evaluate the effects of
Ocrelizumab using both structural and functional MRI. Only the baseline resting-state fMRI data
were included in this thesis research (chapter 3, 25 MS and 41 controls); therefore, the results
described here were independent to pharmaceutical interventions.
35
- The GFM2 data set. These data were acquired in Pacific Parkinson’s Research Centre (PPRC)
and the design was to identify biomarkers for PD and evaluate the effects of vestibular stimuli
interventions using advanced structural MRI, resting-state fMRI, and task-driven fMRI (with
stimulus). Clinical assessments were done in both 24 PD and 15 healthy subjects. The data were
included in chapter 5 and 6.
Since this work is from a pooling of different data sets, some of the cognitive tests are not perfectly
aligned across studies. However, for the studies which assessed cognitive performance, several
domains are commonly evaluated in both disease populations such as executive function,
processing speed, working memory, and attention.
36
Chapter 2: Cognitive profiles in Parkinson’s Disease and Multiple Sclerosis
In this chapter, the cognitive profiles in PD and MS are investigated utilizing a machine-learning
approach – Canonical Correlation Analysis (CCA), which can be seen as an extension of a
multivariate linear regression model. The purpose of the study is to explore the associations
between cognitive domains, demographical characteristics, and clinical data such as diseases
severity and comorbidity. The results shall help understand how the disease affects cognitive
functions in neurological disorders.
2.1 Introduction of cognitive profiles in Parkinson’s Disease and Multiple Sclerosis
Parkinson’s disease (PD) is a neurodegenerative movement disorder resulting in motor symptoms
of tremor, rigidity, bradykinesia, and postural instability. In addition to motor symptoms, non-
motor deficits, especially cognitive impairments, have a major impact on quality of life in patients
with PD. Patients with PD show cognitive deficits in several common domains such as attention,
memory, visuospatial, and executive functions [Pagonabarraga and Kulisevsky, 2012; Poewe,
2008]. These cognitive deficits are typically assessed with such tasks as Digit Span, the Tower of
London Test (ToL), the Trail-Making-Test (TMT), and the Verbal Fluency Task [Miller et al.,
2013; Williams-Gray et al., 2007]. Older age, non-tremor subtype, and higher Unified Parkinson's
Disease Rating Scale (UPDRS) scores are risk factors for the rapid overall cognitive decline
[Williams-Gray et al., 2007], and specifically, information retrieval and visuospatial abilities can
predict global cognitive impairments in PD. Rather than simply categorising patients into
cognition-intact and cognition-impaired subtypes, there is substantial heterogeneity of cognitive
37
performance in PD, and “frontal”, “posterior”, and “mixed” subtypes have been proposed [Miller
et al., 2013]. Related details of cognitive subtypes in PD have been described previously in section
1.1.2. In short, patients with the “frontal subtype” commonly demonstrate executive dysfunction
related to decreased dopamine levels in frontostriatal circuits. Patients with the “posterior subtype”
constantly show visual-spatial problems related to parietal-temporal deficits as well as
degeneration of cholinergic fibers. The disease severity of this subtype can predict dementia and
patients often demonstrate symptoms of postural instability and gait disorder (PIGD). Patients with
the mixed subtype cannot be easily categorized into the above subtypes. As a result, subtypes
proposed in the “dual syndrome hypothesis” are overlapping with the frontal/posterior subtypes
[Kehagia et al., 2012; Miller et al., 2013].
Traditional cognitive tests typically attempt to probe one aspect of cognition (e.g. attention), but it
is very difficult to test components of cognition in an isolated fashion. Hence performance on
different cognitive tests often correlate with each other, and it is likely that novel analyses utilizing
machine-learning approaches will be more suitable in establishing overall cognitive profiles in
subjects with neurological disorders.
Multiple sclerosis (MS) is a neuroinflammatory disease which causes sensory and motor
disturbances, diplopia, ataxia, bladder disturbance, fatigue, and cognitive dysfunction [Gelfand,
2014]. Among these symptoms, cognitive impairments may have a strong impact on subjects’
daily lives as the commonly affected cognitive domains in MS include impaired information
processing speed, attention, memory, and executive function [Wallin et al., 2006]. A characteristic
feature of MS is that more women are affected than men. In addition to prevalence, gender
differences have been addressed in MS with respect to genetic susceptibility, clinical presentation,
38
effects of sex hormones, immune system, and response to therapy [Harbo et al., 2013]. Previous
studies have proposed several links between gender effects and neuroimaging findings such as
cortical atrophy, functional connectivity, and white matter changes [Savettieri et al., 2004;
Schoonheim et al., 2012a; Schoonheim et al., 2012b]. However, despite the clinical importance of
cognition for overall well-being in the MS population, research on the direct association between
gender and cognitive dysfunction has not been widely reported in MS.
Schoonheim et al. demonstrated that male subjects with MS showed significantly lower cognitive
scores compared to male controls in several cognitive domains such as executive functioning,
verbal memory, processing speed, working memory, attention, and psychomotor speed, yet these
domains were relatively preserved in female subjects [Schoonheim et al., 2012b]. The authors also
found that male subjects had a better correlation between subcortical volume and a crude cognitive
marker (the average Z-score of a battery of cognitive tests). In addition, significant white matter
changes such as lower fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity (AD),
and radial diffusivity (RD) in diffusion tensor imaging (DTI) data have been observed in male
subjects who had lower cognitive Z scores, but not in female subjects whose cognition was
relatively intact [Schoonheim et al., 2014]. In fMRI, male subjects demonstrated decreased resting
state functional connectivity as well as lower network efficiency in association with deteriorating
cognitive performance [Schoonheim et al., 2012a], especially reduced visuospatial memory. While
these studies have focused on the relationship between cognitive dysfunction and
functional/structural changes in the brain, less emphasis has been on the characterization of
cognitive differences and gender effects in MS.
39
Several factors appear to be predictors of cognitive dysfunction in MS: early age of onset, male
sex, secondary progressive course, neurodegeneration, and low baseline intelligence [Benedict and
Zivadinov, 2011]. Male subjects are more likely to develop a severe disease course including
physical disability and cognitive impairments. In MS subjects, it appears that cognitive
dysfunction in females is less dependent on factors such as physical disability, while cognitive
decline in males is correlated with Expanded Disability Status Scale (EDSS), age, disease duration,
and education level [Savettieri et al., 2004]. Another study also found that female subjects
performed better on the Mini Mental State Examination (MMSE), Wisconsin Card Sorting Test
(WCST), language, and memory tests in Repeatable Battery for the Assessment of
Neuropsychological Status Update (RBANS) [Beatty and Aupperle, 2002]. Thus, male subjects
appear to be particularly cognitively vulnerable in MS.
In this study, eight neuropsychological tests that employed executive function were administered
to an MS cohort (13 males, 33 females). Since these tests all probe aspects of executive function,
it is likely that performance in these individual tests will be related to (correlated with) one another,
yet each test still likely provide unique information. It is therefore desirable to determine if the
tests could be judiciously combined in some way and if overall performance across the
neuropsychological tests. Specifically, we wanted to establish if there was an established overall
pattern of neuropsychological deficits seen in MS, and if this pattern of deficits could be predicted
based on disease severity and other demographic factors, including gender. Our implicit hypothesis
is that a combination of gender, age, education, and clinical indices will correspond with a
combination of results seen in neuropsychological tests. In order to achieve this, we employed
canonical-correlation analysis (CCA), a type of machine learning method used to identify patterns
40
in large data sets. We show that a combination of demographic (i.e., gender, age, education, alone
or in combination) and clinical indices (i.e., disease duration, severity of disability, affective states)
accurate predicts results of frontal lobe testing. When we then interrogated the particular
combination of demographic factors that predicted neuropsychological test performance in MS,
we found that gender had the greatest influence.
In the PD cohort, the same method was applied to investigate the cognitive profiles and the
relations to demographics in PD, specifically whether gender also contributed to cognitive
differences.
2.2 Materials and methods
2.2.1 Subjects
PD data are from the Parkinson's Progression Marker Initiative (PPMI) and MS data are from
COGMS project.
Thirty-one subjects with PD who enrolled in the PPMI were included in the study. All subjects
underwent neuropsychological assessments and imaging scans with T1-weighted MRI and resting-
state fMRI (rsfMRI). All data in this study were acquired at baseline (fMRI results are reported in
another chapter). The inclusion criteria required subjects must show at least two of the following:
resting tremor, bradykinesia, rigidity or either asymmetric resting tremor or asymmetric
bradykinesia. Subjects had a diagnosis of PD for two years or less, Hoehn and Yahr stage I or II,
off medication, age 30 years or older at diagnosis, and ability to provide written consent. Exclusion
criteria included atypical PD syndromes, taking any PD medication, taking levodopa or dopamine
agonists prior to baseline for more than a total of 60 days, dementia (screened by Montreal
41
Cognitive Assessment (MoCA)), and any other medical or psychiatric condition or lab
abnormalities. Demographics were shown in Table 2.1. For the analysis of cognitive profiles, all
31 subjects were included and three of them showed MCI.
Forty-six relapsing-remitting MS subjects (13 males, 33 females) were enrolled in a study
examining the relations between results of various MRI sequences (including Diffusion Tensor
Imaging, Myelin Water Imaging, lesion load, resting state fMRI functional connectivity, and
cortical thickness), cognitive performance, clinical measures and demographics. For the purposes
of this chapter, we focus on clinical measures and demographics (Table 2.2) as the imaging will
be the subject of another report (in other chapters). All the subjects fulfilled the McDonald criteria
for MS diagnosis and were recruited from the MS clinic at the UBC Hospital. Referrals underwent
a telephone screen and subjects with significant untreated depression and/or other psychiatric
illness, learning disabilities, history of drug or alcohol abuse, had used steroids in the last 3 months
or evidence that they were in an active flare were excluded from this study. Subjects were also
screened for adequate motor skills for their ability to manipulate a pencil which would have direct
impact on some of the tasks (i.e., Coding and Symbol Search). Ethics approval was issued by the
University of British Columbia's Research Ethics Board and all subjects provided signed informed
consent.
Parkinson’s subject (mean±SD)
10 females 21 males
demographics & clinical data
age 59.85±10.8 61.48±9.4
UPDRS 17.3±7.1 14.67±12.9
depression 4.7±1.2 5.43±0.9
education in years 17.1±2.3 16.67±3.2
cognitive scores
42
MOCA 27.8±2.1 27.38±2.1
BJLOTOT 25.2±4.0 25.24±4.0
HVLTTOT 25.2±5.8 22.43±5.2
HVLTDELAY* 9.9±2.4 7.57±2.7
DVT-HVLTTOTAL* 53.3±13.3 40.05±13.1
DVT-HVLTDELAY* 57.5±9.1 45.29±17.4
DVT-HVLTRETENTION 56.6±8.7 49.61±16.1
LNS-RAW 11.0±2.5 9.9±2.7
SFCOM 52.1±11.9 45.10±9.8
SFVEG* 15.6±3.9 12.28±3.7
SFANI 21.0±5.5 21.29±4.9
SFFRU* 15.5±4.3 11.52±2.8
SDMT 41.10±9.8 39.67±9.0
Table 2.1 Demographics, clinical data, and cognitive scores in PD subjects.
[UPDRS = Unified Parkinson's Disease Rating Scale, MOCA = Montreal Cognitive Assessment, BJLOTOT =
Bento Line Orientation Total Score, HVLTTOT = Hopkins Verbal Learning Test-Revised Total Score,
HVLTDELAY = HVLT Delayed Recall Score, DVT-HVLTTOTAL = standardized HVLT Total Score, DVT-
HVLTDELAY = standardized HVLT Delayed Recall Score, DVT-HVLTRETENTION = standardized HVLT
Recognition Trial Score, LNS-RAW = raw Letter-Number Sequencing Test Score, SFCOM = Sematic Fluency
Test – combination, SFVEG = Sematic Fluency Test – vegetable trial, SFANI = Sematic Fluency Test – animal
trial, SFFRU = Sematic Fluency Test – fruit trial, SDMT = Symbol Digit Modalities Test]
* Differences in two-sample t-test with p<0.05
Multiple sclerosis subjects (mean±SD)
33 females 13 males
Demographics & Clinical measures
Age 41.1±10.3 45.8±11.6
Education (years) * 15.3±2.2 13.7±2.7
EDSS 2.3±1.6 1.9±1.5
Disease duration (months) 129.0±86.2 128.9± 133.1
Neuropsychological scores
WAIS Digit Span ScS 9.0±2.5 8.2±2.2
Arithmetic ScS 9.0±2.6 8.9±3.3
Letter Number Sequencing ScS 9.6±2.7 8.8±2.4
Symbol Search ScS 10.4±3.5 9.2±3.2
Coding ScS 10.3±2.6 9.0±2.1
43
2.2.2 Test battery
The neuropsychological assessments performed on PD subjects have been previously described
[Lebedev et al., 2014]. In total, there were 13 scores from the assessments included MoCA, Benton
Judgment of Line Orientation (BJLO), raw Hopkins Verbal Learning Test-Revised (HVLT), raw
Working Memory Index SS 93.7±12.4 91.0±14.7
Processing Speed Index SS 101.9±16.0 94.5±14.1
FAS ADJ 41.0±10.7 37.4±14.4
WCST Errors SS 93.1±13.7 82.2±19.0
Perseverative Responses SS 92.3±12.6 86.2±23.6
Perseverative Errors SS 91.1±12.9 86.3±23.6
Categories Completed Raw *** 5.5 ±1.1 3.3±1.9
Trials to Complete First Category Raw*** 12.7±4.4 38.6±38.3
Failure to Maintain Set Raw 0.5±0.8 1.0±1.6
Learning to Learn Raw N/A N/A
TMT A Raw * 30.7±12.3 40.3±13.3
Z -0.7±1.5 -1.2±1.4
E Raw 0.2 ±0.4 0.1±0.3
TMT B Raw 74.2±61.0 110.7±49.8
Z -0.5±2.3 -1.5±2.5
E Raw 0.4±1.0 0.6±1.1
MDI Total Score Raw 31.9±22.4 38.3±21.8
T 47.3±8.9 52.7±10.2
STAI State SS 49.8±10.4 50.2±9.7
Trait SS 54.6±12.5 58.5±7.8
FSS Raw 38.2±15.2 35.2±14.8
Table 2.2 Demographics, clinical measures, and neurological scores in MS subjects.
[WAIS: Wechsler Adult Intelligence Scale—Fourth Edition, FAS: Verbal Letter Fluency Test, WCST:
Wisconsin Card Sorting Test, TMT A: Trail Making Test A, TMT B: Trial Making Test B, MDI: Multiscore
Depression Inventory, STAI: State-Trait Anxiety Inventory, FSS: Fatigue Severity Scale, ScS: scaled score
adjusted by age, SS: standardized score, ADJ: adjusted for age/gender/education, Raw: raw score, Z: Z score,
E: error, T: t score, N/A: more than 5 missing data sets]
* Differences in two-sample t-test with p<0.05
*** Differences in two-sample t-test with p<0.001
44
HVLT delayed recall, standardized HVLT, standardized HVLT delayed, standardized HVLT
recognition trial, raw Letter-Number Sequencing test (LNS), Sematic Fluency Test – combination
(SFCOM), Sematic Fluency Test – vegetable trial (SFVEG), Sematic Fluency Test – fruit trial
(SFFRU), Sematic Fluency Test – animal trial (SFANI), and Symbol Digit Modalities Test
(SDMT) scores (Table 2.1). MoCA screened for overall cognitive functions, BJLO evaluated
visuospatial function, HVLT assessed verbal memory, executive functions were evaluated by the
Fluency test and MoCA subtests, and SDMT and LNS tested attention. Therefore, memory,
visuospatial, and attention/executive functioning domains were measured in this study, which were
the most common affected domains in PD. Questionnaires for depression were also administered
to measure affective symptoms.
All MS subjects underwent eight psychometric measures assessing processing speed, working
memory, executive function, and attention. This battery was selected not for clinical purposes. The
tests included WAIS IV (Wechsler Adult Intelligence Scale IV) subtests (Digit Span, Arithmetic,
Letter Number Sequencing, Symbol Search, and Coding), Verbal Letter Fluency Test (FAS),
Wisconsin Card Sorting Test (WCST), and Trail Making Test A and B (TMT A and B). Composite
Index scores from the WAIS IV were also obtained including Working Memory Index (WMI)
which utilized scores from Digit Span, Arithmetic, and Letter Number Sequencing subtests. The
WAIS IV Processing Speed Index (PSI) is based on the Symbol Search and Coding subtests.
Clinical questionnaires were also administrated in order to examine affective status, which
included Multiscore Depression Inventory (MDI) and State-Trait Anxiety Inventory (STAI). The
other measure administered was a widely used fatigue measure in the MS field, the Fatigue
Severity Scale (FSS). The duration of the assessment was approximately one hour. Raw cognitive
45
scores were converted to standardized scores using published normative data that considers
psychometric factors such as age, education and gender. The patient’s disability rated by the
Kurtzke Expanded Disability Status Scale (EDSS) was determined by a neurologist at the time of
recruitment or scanning.
Although the test batteries were different between PD and MS groups, some sub-tests were the
same. The Letter Number Sequencing (LNS) and the Verbal Fluency Test (SFCOM in PD, FAS
in MS) were included in both test batteries. In addition, the Symbol Digit Modality Test (SDMT)
in PD was very similar to the Coding sub-test in MS and they both evaluated processing speed.
2.2.3 Analysis
Two-sample t-tests (two-tailed) were carried out with all the scores in male and female groups.
Transformed scores were calculated in R (version 3.2.0)- which is a language and environment for
statistical computing- if normality did not hold. CCA [Krzanowski, 1988] was done in MATLAB
(The MathWorks, Inc.) to determine if demographics were related to cognitive scores. CCA was
chosen to study the multivariate data and it can be considered one of the original “machine
learning” algorithms that are now used to explore “big data” in many fields [Correa et al., 2008;
Donner et al., 2006; Palmer, 1993]. The advantage of the CCA method is that it is an extension of
regression that enables additional factors to be included. If we have two sets of correlated variables
(e.g the performance on the neuropsychological tests will likely be correlated, and some
demographic factors such as disease duration and age will likely be correlated), CCA attempts to
find the linear combinations of these two sets of variables that maximally correlate with each other.
Note that this will likely be a more powerful approach than regression where one could try and
46
predict the neuropsychological performance on 1 test – i.e. in regression one only has 1 dependent
variable.
In PD, the two sets were the cognitive scores measured by neuropsychological assessments and
the demographics of 31 PD subjects. The demographics included gender (encoded as female 2 and
male 1), age, UPDRS scores, Depression, and years of education. The cognitive scores included
MoCA, BJLO, raw HVLT, raw HVLT delayed recall, standardized HVLT, standardized HVLT
delayed, standardized HVLT recognition trial, raw LNS, SFCOM, and SDMT scores. As in
sematic fluency, the combination trial should represent the performance of sematic fluency in
animal, vegetable, and fruit trials, so we only included SFCOM instead of taking all 4 scores. All
scores were normalized into z-scores before CCA. We performed CCA in leave-one-out fashion,
to ensure robustness of our results, and reported the loadings (i.e. the correlation between
transformed CCA data and raw scores) for each variable. Specifically, we excluded one subject at
a time and performed the CCA analysis each time. The variability in the weightings/loadings, when
each of the subjects was removed, was recorded. This gives an estimate of how much the results
could change if more subjects were added. If the 95% confidence intervals based on the cross-
validation of each variable did not cross zero, we regarded this variable as a significant contributor
to the CCA model.
In MS, in keeping with prior neuropsychological approaches, we analyzed timed and untimed tests
separately. We performed CCA on cognitive scores divided into two groups (timed and untimed
tests) and demographical variables. The “timed group” included 4 affective variables (Multiscore
Depression Inventory Total T scores, State-Trait Anxiety Inventory standardized State scores and
Trait scores, and Fatigue Severity Scale Raw scores) and 8 variables from the tests in which
47
subjects were timed during performance, including 3 scaled scores in WAIS-IV (Arithmetic,
Symbol Search, and Coding), Verbal Letter Fluency Test, and 4 measures of Trail-Making Tests (
transformed TMT A scores, TMT A Z scores, transformed TMT B scores, and TMT B Z scores).
Due to non-normality, raw scores of TMT A and B were transformed by taking the square root of
the reciprocal of the raw scores (the transformed score could be expressed as the
following:√1
𝑇𝑀𝑇 𝑟𝑎𝑤 𝑠𝑐𝑜𝑟𝑒𝑠). In the “untimed group”, 4 affective variables (same as “timed group”)
and 8 cognitive variables from the tests in which subjects were not timed were included. The
cognitive 8 variables were 2 scaled scores in WAIS-IV (Digit Span and Letter Number
Sequencing) and 6 measures in WCST (standard score of Errors, Perseverative Response,
Perseverative Errors, raw scores of Categories Completed, Trials to Complete First Category, and
Failure to Maintain Set). Age, gender, education, EDSS scores, and disease duration were included
as demographical variables in both groups. Of note, due to the data and sample size, we did not
perform analysis in different groups in PD.
Calculations were performed in leave-one-out fashion to control for multiple comparisons and
ensure the robustness of our results as previously described. The error bars indicated the variability
of the results across subjects.
2.3 Results
2.3.1 PD
In PD, CCA revealed that demographics/clinical data were inter-correlated with cognitive scores
(Figure 2.1). In the combination of demographics and clinical data, gender, age, depression, and
48
education all showed significant canonical loadings, but UPDRS scores did not have an impact on
this model. In the combination of cognitive scores, all variables demonstrated significant loadings
except the standardized HVLT Recognition Trial Score. Variables with loadings the same sign
were correlated with each other (i.e. negative loadings in demographics were positively correlated
with negative loadings in cognitive scores) and anti-correlated with the variables that demonstrated
the opposite sign. The two linear combinations of these two sets of variables were significantly
correlated with each other with a correlation coefficient of 0.90 and p=0.011.
The model demonstrated gender differences of cognitive performances in PD, where female
subjects with more years of education were related to higher scores in almost all cognitive tests as
they showed higher performance z-scores than males (Appendix A.1). Further analyses supported
this observation, as the mean z-scores of all cognitive tests/variables with significant CCA loadings
were higher in female subjects than male subjects except the BJLO test (Appendix A.1). Therefore,
the variable gender in the model (shown in Figure 2.1) demonstrated an overall common effect to
cognition across tests rather than specific patterns in individual tests. Similarly, education
supported overall cognitive functions in PD. Meanwhile, higher age and higher scores in
depression scale were anti-correlated with all cognitive scores which showed significant loadings,
implying that aging and PD comorbidities such as depression worsened cognitive abilities.
49
Figure 2.1 The CCA model shows inter-correlated cognitive patterns with demographics and clinical data in
PD. The linear combination of demographics/clinical data is significantly correlated with the linear
combination of cognitive performances with correlation coefficient 0.90 and p 0.01 (left panel, cross-
validation with leave-one-out). Gender, age, depression, and education have significant loadings in the
combination of demographics and clinical data (middle panel, i.e. 95% confidence interval does not cross
zero as indicated by error bars). In the combination of cognitive scores (right panel), all the variables show
significant loadings except standardized HVLT Recognition Trial Score as the 95% confidence interval
crosses zero. [GEN: gender, EDUY: education in years, MoCA: Montreal Cognitive Assessment, BJLOTOT
= Bento Line Orientation Total Score, HVLTTOT = Hopkins Verbal Learning Test-Revised Total Score,
HVLTDELAY = HVLT Delayed Recall Score, DVT-HVLTTOTAL = standardized HVLT Total Score,
DVT-HVLTDELAY = standardized HVLT Delayed Recall Score, DVT-HVLTRETENTION = standardized
HVLT Recognition Trial Score, LNS-RAW = raw Letter-Number Sequencing Test Score, SFCOM = Sematic
Fluency Test – combination, SDMTTOT = Symbol Digit Modalities Test total scores]
48
2.3.2 MS
In MS, CCA results showed that gender (mean canonical coefficient: timed = 1.01, untimed =
0.94) appears to be an important factor (i.e. weightings can be distinguished from zero) in the
combination of demographics for both timed and untimed tests. In the untimed group, EDSS (mean
canonical coefficient: 0.26) was influential in addition to gender. Age, education, and disease
duration showed limited effects on both timed and untimed tests (mean canonical coefficient:
timed/untimed = 0.07/0.00, 0.03/-0.05, -0.02/-0.03, respectively) (Figures 2.2&2.3).
Since we assigned gender male as 1 and female as 0 in the calculation, the variables which showed
positive canonical coefficient indicated higher scores in male subjects as long as gender also
showed positive weightings. On the other hand, the tests on which female subjects obtained higher
scores show negative canonical coefficients. This was because the model treated female gender
(0) as “baseline”. The weightings of gender indicate the effects of gender male compared to female
in combination of demographics as well as combination of cognitive scores. In the “timed group”,
transformed TMT A and B scores showed strongly negative effects compared to other variables
(mean canonical coefficient: -28.3 and -31.8, respectively) (Figure 2.2). In contrast, the TMT A Z
score demonstrated a positive mean canonical coefficient with 0.64 in the combination of cognitive
scores, illustrating that there was a gender-specific pattern in timed cognitive tests where female
subjects had higher transformed TMT A and B scores and male subjects had higher TMT A Z
scores. In addition, the combination of cognitive scores and demographics were highly correlated
with each other (correlation r = 0.84, p-value < 0.001, Figure 2.2 right panel), meaning that the
linear combination of all the variables in demographics and all the variables in timed cognitive
scores forms a significant model to explain and predict the data.
49
Figure 2.2 CCA results of timed cognitive tests in MS. The left panel shows that among five demographical
variables, gender has a relatively strong effect in the combination. The middle panel illustrates the
weightings of every variable in the combination of timed cognitive scores based on their association with
demographics (negative scores indicate females performing test better). TMT performances show strong
effects (mean canonical coefficient: transformed TMT A and B = –28.3 and–31.80, respectively). The right
panel demonstrates the high correlation (r=0.84) between two sets of multivector (demographics and timed
cognitive scores) with significance (p=0.0002). The error bars indicate how various the results are during the
leave-one-out calculations.
[AGE: age in years, GEN: gender, ED: education in years, EDSS: Expanded Disability Status Scale, DDyear:
disease duration in years, AR.ScS: scaled WAIS Arithmetic, SS.ScS: standardized Symbol Search, CD.ScS:
standardized Symbol Coding, FAS.ADJ: adjusted Verbal Letter Fluency, sq.recip.TMT.A.Raw:
transformed Trail-Making Test, Part A scores, TMT.A.Z: Trail-Making Test, Part A Z scores,
sq.recip.TMT.B.Raw: transformed Trail-Making Test, Part B scores, TMT.B.Z: Trail-Making Test, Part B
Z scores, MDI.TOT.T: Multiscore Depression Inventory Total T scores, STAI.S.SS: standardized State-
Trait Anxiety Inventory State scores, STAI.T.SS: standardized State-Trait Anxiety Inventory Trait scores,
FSS.Raw: Fatigue Severity Scale Raw scores.]
50
Figure 2.3 shows the results from the “untimed group”. Male subjects had higher scores (i.e. these
scores showed higher weightings in male subjects) on WCST Errors, WCST Perseverative Errors,
Trials to Complete First Category, Trait anxiety, and fatigue (mean canonical coefficient: 0.03,
0.14, 0.01, 0.02, and 0.01, respectively), while female subjects obtained higher scores (i.e. these
scores were more influential on the linear model of all scores in female subjects) on WAIS IV
Digit Span, WAIS IV Letter Number Sequencing, WCST Perseverative Responses, WCST
Categories Completed, Failure to Maintain Set, and state anxiety (mean canonical coefficient: -
0.03, -0.04, -0.14, -0.29, -0.06, and -0.01, respectively). This implied that female subjects
completed more categories on the WCST than males, but higher Perseverative Responses also
indicated that female subjects were potentially more prone to cognitive inflexibility as they
changed strategies less frequently. Moreover, female subjects also did better on the Digit Span and
Letter-Numbering Sequencing tasks, meaning that they had better attentional abilities since the
scores of these two tests partially form Working Memory Index. Finally, the two sets of variables
(combination of cognitive scores and combination of demographics) were highly correlated with
each other in untimed group (correlation r = 0.84, p-value < 0.001, Figure 2.3 right panel),
indicating that the linear combination of all the variables in demographics and untimed cognitive
scores forms a model explaining the data.
51
Figure 2.3 CCA results of untimed cognitive tests in MS. The left panel shows that gender and EDSS have
strong effects in the combination of demographics compared with other variables. The middle panel
illustrates the weightings of every variable in the combination of untimed cognitive scores. WCST Errors
showed weightings in male subjects as well as WCST Perseverative Errors, transit anxiety, fatigue, and
WCST Trials to Complete First Category (mean canonical coefficient: 0.03, 0.14, 0.02, 0.01, and 0.01,
respectively). WAIS IV Digit Span, WAIS IV Letter-Number Sequencing, WCST Perseverative Responses,
WCST Categories Completed, Failure to Maintain Set, and state anxiety demonstrated effects in female
subjects (mean canonical coefficient: –0.03, –0.04, –0.14, –0.29, –0.06, and –0.01, respectively). The right
panel demonstrates that these two sets of multivector (demographics and timed cognitive scores) are
significantly correlated with each other (p=0.001, r=0.84). The errorbars indicate how various the results
are during the leave-one-out calculations.
[AGE: age in years, GEN: gender, ED: education in years, EDSS: the Expanded Disability Status Scale,
DDyear: disease duration in years, WAIS.DS.ScS: scaled WAIS Digit Span, LN.ScS: scaled WAIS Letter
Number Sequencing, WCST.TE.SS: scaled WCST Total Errors, WCST.PR.SS: standardized WCST
Perserverative Responses, WCST.PE.SS: standardized WCST Perserverative Errors, WCST.CC.Raw: raw
WCST Categories Completed, WCST.TCFC.Raw: raw scores of WCST Trails to Complete First Category,
WCST.FMS.Raw: WCST Failure to Maintain Set, MDI.TOT.T: Multiscore Depression Inventory total T
scores, STAI.S.SS: standardized State-Trait Anxiety Inventory State scores, STAI.T.SS: standardized State-
Trait Anxiety Inventory Trait scores, FSS.Raw: Fatigue Severity Scale Raw scores.]
52
2.4 Discussion
Several studies have utilized CCA to study inter-correlations between behavioural data, cognition,
imaging findings, and daily activity in both healthy subjects and disease populations [Alonso Recio
et al., 2013; Davis et al., 2011; Lin et al., 2017; Nilsson et al., 2016; Perry et al., 2017; Smith et
al., 2015], indicating that CCA is a useful tool to study relations between two sets of data.
With CCA method, we reported that 1) there were gender specific cognitive patterns in MS and 2)
some demographical variables had stronger effects on cognitive performance. More importantly,
utilizing a focused test battery of tests sensitive to aspects of executive functioning, there were
significant gender differences. The findings from the current study endorse the need for sensitivity
to include both untimed and timed tasks in the cognitive assessment of individuals with remitting
and relapsing MS. Similarly, CCA models also revealed the inter-correlated behaviour between
demographics, clinical measures, and cognitive performances that are commonly impaired in PD
including executive function in this study. Interestingly, we also observed gender effects in PD
and some demographical data showed stronger impacts than others.
2.4.1 Cognitive profiles in PD: female gender and education support better cognitive
function, while comorbidity is related to poor cognition
In PD, a CCA approach found that working memory, attention, planning, and problem solving
were inter-correlated with visuospatial memory and episodic memory in early stage PD and
executive function (index of working memory, attention, planning) and visuospatial memory
contributed the most to cognitive deficits [Alonso Recio et al., 2013]. Although these studies have
shown that there are inter-correlations in cognition and disease, gender differences have not been
53
previously observed in PD with a CCA approach. We found that education supported better
cognitive performance in visuospatial, verbal learning, verbal memory, working memory,
executive domains as well as information retrieval abilities. Our results are consistent with the
concept of cognitive reserve theory, whereby education is one of the factors that may enhance
neuronal plasticity to maximize/optimize performance or strengthen the ability to engage altered
brain networks in the setting of regular aging in healthy individuals and/or neurodegenerative
disease [Stern, 2002; Tucker-Drob et al., 2011; Vance et al., 2010]. The enhanced plasticity in a
variety of neural circuits includes increased dendritic connections between neurons, an increase in
neutrophic factors, and stronger connections between neurons, which provides more neural
resources supporting brain function and increases the capacity to cope with damages due to aging,
injury, and diseases [Ansari, 2012; Vance et al., 2010]. Therefore, subjects can tolerate more
insults while still being able to perform normally or even more efficiently in cognitive tests before
functional impairments are apparent. As a result, the factors which trigger such alterations can be
considered protective factors such as education in our study [Stern, 2002]. However, longitudinal
studies have suggested that aging may be a stronger factor regarding cognitive changes than
education [Tucker-Drob et al., 2011; Zahodne et al., 2011]. This is similar to what we found where
both education and age showed strong loadings in the CCA model -- a pattern that has been
reported in healthy subjects as well [Perry et al., 2017].
The gender differences that we discovered have been observed previously in PD, but the
differences in cognition have not been extensively studied and results of non-motor symptoms
have been inconsistent [Augustine et al., 2015; Miller and Cronin-Golomb, 2010]. In general,
female subjects tend to perform better on many cognitive tests in older healthy subjects [Smith et
54
al., 2015]. However, a study with 1700 individuals with PD reported that motor-symptoms were
not different between male and female, but female subjects fared better in non-motor aspects which
included SDMT performance and daily functioning [Augustine et al., 2015]. Moreover, male
subjects have faster decline of cognitive abilities in verbal, letter, and category fluency tests; while
females tend to have worse visuospatial abilities [Corkin et al., 2003; Riedel et al., 2008]. Our
results also demonstrated that females performed better in SDMT, verbal fluency tests, and overall
cognition measured by MoCA, but presented with worse visuospatial function (Appendix A.1),
consistent with previous research. Generally, attention, executive functioning, and overall
cognitive function is relatively less affected in female PD subjects, which is possibility supported
by education. The purported mechanisms of gender differences in cognition are unclear, but it has
been postulated that estrogen may impart neuroprotective effects by activating receptors, altering
protein expression, and activating kinase on dopaminergic pathways across cortices (e.g. the
hippocampus and prefrontal cortex) in several neurodegenerative disorders [Brann et al., 2007;
Green and Simpkins, 2000; Miller and Cronin-Golomb, 2010]. Moreover, estrogen therapy may
prevent cognitive decline and support the abovementioned viewpoints, but the effects vary across
lifetime [Sherwin, 2012].
Unsurprisingly, in PD, we found that depression had a negative effect on cognitive function,
consistent with previous research [Starkstein et al., 1989]; however, we did not have evidence to
probe whether cognitive decline induced depression or depression manifested cognitive
dysfunction in PD – an area of debate [Menza et al., 1993]. We can only interpret that there is a
strong association between the co-existence of aging and psychiatric comorbidity and poor
cognitive performance that requires attention and executive abilities. Surprisingly, the commonly
55
used measure of overall disease severity, UPDRS, did not show significant loadings, which
implied that overall motor assessment did not directly reflect cognitive decline. Motor deficits and
cognitive dysfunction engage partially different pathways and systems [Sawamoto et al., 2002],
but there is still a certain degree of association of degeneration as individuals with higher disease
severity have a higher chance of developing cognitive impairments.
2.4.2 Cognitive profiles in MS: female gender supports better cognitive function and
disease severity is related to worse performance
In MS, gender differences have been reported in previous studies which showed that females
perform better than males on memory tasks and the Wisconsin Card Sorting Test [Beatty and
Aupperle, 2002; Schoonheim et al., 2012a]. Moreover, male sex has been speculated as a risk
factor for development of severe cognitive decline [Benedict and Zivadinov, 2011]. These studies
investigated cognitive function based on the responses on cognitive tests in male and female
separately. Given that cognition is a complex multi-dimensional entity which can be assessed
within different domains [Heyes, 2012], it is reasonable to assume that there are multiple
intercorrelated variables modulating cognition. Therefore, we investigated the intercorrelation
from combinations of demographics and cognitive variables through CCA, determining which
linear combination of demographics and cognitive scores best represent the data in two genders as
a whole.
As figure 2.2 and 2.3 demonstrate, gender was a strong demographic factor in timed tests and both
gender and EDSS were influential in untimed tests as the rest of the variables had limited effects.
Demographical factors were taken into account in our model, which highlighted the fact that both
EDSS and gender were influential factors for cognition in our data especially gender. Our results
56
were in agreement with previous research showing that gender was a predictor for cognitive
dysfunction in MS and male subjects generally showed worse performance [Benedict and
Zivadinov, 2011; Schoonheim et al., 2012b]. Education and affective states can directly impact
cognition [Bruce and Arnett, 2005]. However, in this study education did not impact cognitive
profile even though female subjects had higher education. Moreover, none of the affective
variables were significantly different between the two genders and all of them had limited
weightings in CCA results. Although there were trends that the females in this study were more
educated, it was not a significant difference (Table 2.2). Therefore, we concluded that education
and affective status did not significantly influence cognition in our cohort.
However, of note, when rsFC, demographics, and cognitive scores are combined together in
another analysis with a similar approach, education appears to be influential. This study is reported
in chapter 6.
In our MS data, due to the fact that transformed TMT scores were the square roots of reciprocal
TMT raw scores (which had opposite meanings to raw TMT scores), our interpretation of high
weightings in TMT scores (Figure 2.2) is that female subjects performed better on both TMT A
and B tests (i.e. faster responses). We conclude that set-shifting abilities assessed by TMT A/B
tests were less impaired in female MS subjects than males. Finally, the high correlation between
the combination of demographics and combination of timed cognitive scores illustrated that these
linear combinations were significant and robust enough to explain the data. Figure 2.3
demonstrates that gender had strong weightings in the combination of demographics and untimed
scores. Among all the untimed scores, male subjects obtained high weightings on WCST
Perseverative Errors and Errors, illustrating poor performance in the Wisconsin Card Sorting Test.
57
In contrast, female subjects tended to perform Wisconsin Card Sorting Test better because they
had higher scores on WCST Categories Completed. However, they also had higher scores on
WCST Perseverative Responses and Failure to Maintain Set, which possibly indicates an inability
to use feedback to modify their response behaviour. It is interesting to know that perseverative
behaviour, whether errors or responses, are found in both genders. This is more indicative of the
neuropathology (i.e. white matter damages reduce neuronal communication) seen in MS. In
addition, female subjects obtained higher scores on WAIS IV Digit Span and Letter-Number
Sequencing tests, indicating that, compared to male subjects, their basic ability to sustain attention
was less affected. Finally, the untimed group also demonstrated high correlations between the
combination of cognitive scores and combination of demographics, and again, gender and a subset
of WCST scores were the strong factors which explained cognitive patterns.
Gender differences on cognitive tasks have been long established [Miller and Halpern, 2014].
Separating the timed from non-timed tasks is an important, growing trend in the analysis of
neuropsychological test data in multiple sclerosis [Leavitt et al., 2014]. We also analyzed timed
and untimed scores together, but the model was not significant. The current results indicate that
analyzing timed and untimed scores separately is more capable of distinguishing cognitive profiles
between male and female MS subjects.
2.4.3 Commonalities and differences
To conclude, the cognitive profiles in PD and MS shared similar patterns that gender appeared to
be an influential factor in cognition. In both cohorts, female subjects performed cognitive tests
better, especially tests that required set-shifting abilities. On the other hand, there are several
differences in their cognitive profiles. First, some demographical factors showed distinct effects,
58
whereby age and education were influential on cognition in PD but not in MS, which possibly due
to that PD subjects are generally older so aging could impact cognition more than that in MS.
Second, disease severity also showed different effects to cognition. UPDRS was not a significant
factor in PD, while EDSS was associated with worse performance in MS, which raises debates
whether 1) disease severity measured with motor symptoms can represent cognitive impairments
or 2) psychological disabilities show certain degree of associations between clinical symptoms and
cognitive dysfunction as patients with stronger psychological impairments are more likely to
develop cognitive deficits. Further research with a bigger sample size is needed regarding these
issues. Finally, psychiatric comorbidity significantly impacted cognition in PD but not in MS. The
discrepancies may be related to the neurotransmitter dysregularization as several neurochemistry
pathways are altered in PD including serotonin which is highly related to mood disorders [Dauer
and Przedborski, 2003], but similar research is relatively limited in MS. A few studies have
demonstrated that altered neurotransmitter systems in MS are more related to immune function
rather than affective state [Polak et al., 2011], which could be the possible explanation why
psychiatric comorbidities were not significantly associated with cognitive performance in our
results. However, more research is needed to draw firm conclusions. Finally, as PD subjects are
generally older than MS subjects, differences could be partly explained due to age. Aging has been
shown to impact several cognitive domains such as processing speed, working memory, and
executive function by potentially impairing microstructural changes [Fjell et al., 2017; Murman,
2015]. In addition, age-related diseases accelerate the progression of neural loss, neuronal
dysfunction, and cognitive decline [Murman, 2015]. Taken together, the associations between
aging and cognitive decline explained why age demonstrated significant effects to cognition in
PD, an older population, but not MS, a much younger population.
59
2.4.4 Limitations
As described roughly in section 2.2, test batteries used in PD and MS were different even though
some sub-tests were the same. These test batteries were designed to reflect the cognitive
impairments seen in PD and MS. Given that the profiles of cognitive impairments are not exactly
the same, test batteries would most likely vary. However, the common affected domain – executive
function – was evaluated in both batteries with the Verbal Fluency Test, Wisconsin Card Sorting
Test, and Trial-Making-Tests. In addition, as the purpose of the chapter was to explore the
cognitive profiles in PD and MS, healthy subjects were not included in the analysis, which may
fail to establish whether performance was objectively deficient. Finally, although previous
research has implied that the affected neurotransmitter systems in PD are more related to mood
disorders and such links are not seen in MS, further investigations are required to study the
associations between cognition and psychiatric comorbidities in neurological conditions.
2.5 Conclusion
In this chapter, we applied a machine-learning method to explore the relations between cognitive
performance and demographics/clinical data in PD and MS. In PD, the CCA model demonstrates
that female sex and education supported the cognitive performance which requires attention and
executive functioning. Moreover, age and depression were associated with poorer performance,
indicating an association between PD comorbidity and cognitive decline. In MS, we conclude that
gender is one of the influential factors on cognitive performances in subjects with MS. There are
specific cognitive patterns in MS subjects: First, female subjects performed TMT tests better than
males. Second, in untimed tests, male subjects made more errors and female subjects performed
better on the Wisconsin Card Sorting Test. Finally, our results imply that with this particular test
60
battery, female subjects with MS were less cognitively impaired than male subjects. Our study is
unique in that we have discovered gender differences exist on selected executive tasks and report
a trend that gender differences may also exist on timed tasks vs untimed tasks, which has become
an important area of focus in the cognitive assessment of individuals with remitting and relapsing
MS.
Taken together, we propose an inter-correlated pattern between cognitive performance, clinical
evaluation, and demographical characteristics in PD and MS. Moreover, we conclude that gender
is an influential factor to such complex cognitive patterns and female gender shields cognitive
functions regardless pathologies, possibly owning to the neuroprotective effects of estrogen. We
also revealed differences in the cognitive profiles, whereby demographical characteristics, disease
severity, and comorbidities impacted cognition differently in PD and MS.
61
Chapter 3: Resting-state functional connections and cognition
As Chapter 1 hypothesizes, long-range connections appear to be particularly important for
executive function and alterations in these connections might be a common cause of cognitive
impairments in both Parkinson’s Disease (PD) and Multiple Sclerosis (MS). In this chapter, several
strategies are applied to investigate whether neurological diseases demonstrate alterations in long-
range connections as well as other important connections such as interhemispheric ones.
Regression analyses are also used to explore the associations between rsFC and cognitive scores.
3.1 Introduction
3.1.1 Effects of interhemispheric and long-range connections to cognition
Interhemispheric connectivity appears to be particularly important for performance of information
processing and other cognitive functions [Bloom and Hynd, 2005]. Diseases that result in abnormal
volumes of the corpus callosum and/or abnormal interhemispheric connectivity show deficits in
several cognitive domains, such as memory, speed of processing, and executive function [Bodini
et al., 2013; Lee et al., 2010; Mwansisya et al., 2013; Paul et al., 2007; Saar-Ashkenazy et al.,
2014]. For instance, patients with agenesis of the corpus callosum (AgCC) have severely impaired
high-order cognition (e.g. complex thinking, reasoning and decision making, cognitive flexibility)
[Paul et al., 2007]. Decreased interhemispheric connectivity in schizophrenia correlates with
negative symptoms and relates to lower information processing speed [Mwansisya et al., 2013].
Moreover, patients with Alzheimer’s disease and mild cognitive impairment (MCI) have reduced
62
volume in corpus callosum [Lee et al., 2010], and impaired interhemispheric connectivity in post-
traumatic stress disorder is also highly associated with poor performance in associative memory
tasks [Saar-Ashkenazy et al., 2014]. With rsfMRI, patients with major depression disorder (MDD)
show negative correlation between rsFC of homotopic regions and the severity of cognitive
impairments [Wang et al., 2013]. These regions are located in prefrontal areas and cerebellum.
Furthermore, decreased interhemispheric connectivity in stroke patients has been shown to be able
to predict behaviour in multiple domains such as verbal and visual memory [Siegel et al., 2016].
However, only a few of these studies specifically target interhemispheric functional connectivity.
Most of the studies either assess interhemispheric connections with structural data or simply
interpret observed rsFC changes as interhemispheric connectivity.
Long-range connections (i.e. corticocortical connections) have been thought of as supporting
diverse cognitive functions, especially higher-order cognition, as well as resting-state patterns
[Park and Friston, 2013]. Compared to attention, memory ability engages stronger long-range
intrahemispheric connectivity [Hermundstad et al., 2013], showing that higher cognitive load
requires stronger long-range connectivity. Language processing is associated with long-range
connectivity between the inferior frontal gyrus, posterior middle temporal gyrus, basal temporal
cortex, supramarginal gyrus, and superior temporal cortex [Salmelin and Kujala, 2006]. These
long-range connections are affected in diseased populations. For example, long-range connectivity
between the fusiform gyrus and other remote cortical regions is decreased in autism spectrum
disorder (ADS) [Khan et al., 2013]. Decreased connectivity in the fronto-parietal network, perhaps
one of the most well-known cognitive networks that also requires long-range connectivity, has
been reported in a variety of diseases and associated with cognitive deficits such as schizophrenia,
63
chronic traumatic brain injury, and Parkinson’s disease [Han et al., 2016; Sheffield et al., 2015;
Vervoort et al., 2016].
3.1.2 Interhemispheric and long-range connections in Parkinson’s Disease and Multiple
Sclerosis
We hypothesize that interhemispheric connectivity may also be important in maintaining cognitive
performance in MS. Structural research suggests that white matter lesions have a propensity to
involve callosal connections [Bodini et al., 2013], and callosal atrophy is related to lower
processing speed abilities measured by the Symbol Digit Modalities Test (SDMT) at both baseline
and follow-up in MS [Bergendal et al., 2013]. In addition to altered structural interhemispheric
connectivity, studies of functional interhemispheric connectivity with other modalities have also
been described in MS. Decreased magnetoencephalography (MEG) synchronization between the
two hemispheres has been found in MS, suggesting that weakened functional interhemispheric
connectivity could be a biomarker for cognitive impairment [Cover et al., 2006]. Therefore,
studying the relationship between cognitive performance and interhemispheric connectivity is
crucial for further understanding the cognitive profile of MS.
In PD, voxel-mirrored homotopic connectivity (VMHC), which assesses pairwise correlation
coefficient between homotopic regions, has been primarily used to evaluate interhemispheric
connectivity with fMRI data. PD patients with depression demonstrated lower interhemispheric
rsFC in the dorsolateral prefrontal cortex and calcarine, and was associated with overall cognition
measured by the Mini Mental State Examination (MMSE) [Zhu et al., 2016]; compared to healthy
subjects, lower VMHC values in the putamen, sensorimotor cortex, and supramarginal gyrus in
PD subjects are correlated with disease severity and disease duration [Luo et al., 2015b]. In
64
addition, with diffusion weighted MRI data, which quantifies white matter microstructural
integrity, PD subjects exhibited lower white matter integrity of interhemispheric tracts in the
somatosensory cortex, primary motor cortex, supplementary motor area (SMA), and pre SMA
[Fling et al., 2016]; however, the decreased callosal integrity in PD was only correlated with motor
performance rather than cognitive evaluation.
Studies which report the associations between executive function and long-range connections of
rsfMRI in PD and MS have been discussed previously (see Section 1.4).
3.1.3 Analyses
Technical issues may complicate interpretation of functional interhemispheric connectivity. For
interhemispheric connections, similarity between neural activities in homologous regions may not
necessarily be based on transcallosal activity, but rather both hemispheres may be influenced by
common brainstem and/or subcortical input. Functional connectivity measures that distinguish
between direct and indirect (common) influences between homologous regions are therefore
desirable for functional interhemispheric connectivity studies. This can be achieved by partial
correlation, whereby the relation between homologous pairs is assessed controlling for the activity
in another region. Although the VMHC approach, as discussed previously, specifically examines
functional interhemispheric connectivity, pairwise correlation appears to be less robust to noise
and possible effects from other regions compared to partial correlation as used here [Smith et al.,
2013]. For assessing changes in long-range connections, perhaps exploratory analyses are more
suitable given that there is no clear a priori knowledge and definition of such connections.
Although most studies interpret corticocortical connectivity as long-range connections between
65
frontal, parietal, and occipital regions [Park and Friston, 2013; Salmelin and Kujala, 2006], other
connections may have functional significance.
In this study, we examined functional interhemispheric connectivity in MS and PD during resting-
state fMRI using both simple and partial correlations. We then applied elastic net regression
models to investigate the relationship between observed functional connectivity changes and
performance on cognitive tests. Whole brain connectivity patterns were calculated as well to
investigate potential changes in long-range connections. We demonstrate that functional
connectivity features are complementarily associated with performance on cognitive tests.
3.2 Materials and methods
3.2.1 Subjects
PD data are from BCT research project and MS data are from OPERA clinical trial, which is
different from the previous chapter.
We recruited 25 patients with relapsing-remitting MS (mean age ± SD = 37.2 ± 9.5; 10 male, 15
female) and 41 age-/gender-matched healthy control subjects (HC) (mean age ± SD = 34.9 ± 10.1;
14 male, 27 female) in a Phase III randomized, double-blind, double-dummy trial to evaluate the
efficacy and safety of ocrelizumab versus interferon beta-1a in relapsing forms of MS (OPERA II;
NCT01412333). The data included in this study (clinical scores and imaging data) were from the
baseline time point only. Ethics approval was received from the University of British Columbia's
Clinical Research Ethics Board and all subjects provided written, informed consent. HC did not
have psychiatric, medical, cognitive, or other conditions that caused an inability to participate in
an MRI study. Patients had Expanded Disability Status Scale (EDSS) scores ranging between 0
66
and 4 (median EDSS = 2). All patients performed the multiple sclerosis functional composite
(MSFC) battery including the Paced Auditory Serial Addition Test (PASAT). They also performed
a Low Contrast Visual Acuity Test (LCVA), and the Symbol Digit Modalities Test (SDMT) (Table
3.1).
Twelve subjects with idiopathic PD (60.0±9.9 y/o) and 10 age-matched HC were recruited through
the movement disorder clinic at UBC hospital. Ethics approval was issued by the University of
British Columbia's Clinical Research Ethics Board and all subjects provided written, informed
consent. PD subjects were asked not to take their medication one night before the visit (i.e. off
medication). After the first MRI scan and clinical examination in the morning, they took
medication and underwent the second MRI scan in the afternoon on the same day. In addition to
the Unified Parkinson's Disease Rating Scale (UPDRS) part III, questionnaires of apathy and Beck
depression scale were administered and overall cognitive function was assessed by Montreal
Cognitive Assessment (MoCA). Detailed demographics are shown in Table 3.2.
HC subjects (mean ± SD) MS patients (mean ± SD)
Age 34.85 ± 10.1 37.20 ± 9.5
Gender 14 males/27 females 10 males/15 females
EDSS ND 2.14 ± 0.95
Disease duration (months) ND 67.08 ± 64.85
PASAT ND 42.44 ± 15.38
SDMT ND 49.88 ± 11.03
LCVA ND 40.92 ± 9.65
Table 3.1 Demographics of MS and HC subjects.
[HC: healthy control; MS: multiple sclerosis; EDSS: Expanded Disability Status Scale; LCVA: Low Contrast
Visual Acuity; PASAT: Paced Auditory Serial Addition Test-3 seconds; SDMT: Symbol Digit Modalities Test;
ND: no data]
HC subjects (mean ± SD)
mean ± SD
PD subjects (mean ± SD)
mean ± SD
67
age 59.4±6.1 60.0±9.9
sex 6 males/4 females 9 males/3 females
UPDRS ND 29.5±9.6
bradykinesia ND 10.75±3.7
tremor ND 5.25±3.5
rigidity ND 5.92±2.5
PIGD ND 4.58±2.2
H & Y ND 2.5±0.4
disease duration ND 5.92±3.5
medication dosage
(levodopa)
ND 625±351.9
apathy+ ND 7.5±5.4
Beck depression scale+ ND 5.75±5.5
MoCA+ ND 27.14±3.4
Table 3.2 Demographics of PD and HC subjects.
[HC: healthy control; PD: Parkinson’s disease; UPDRS: Unified Parkinson's Disease Rating Scale; PIGD:
Postural Instability and Gait Disorder; H&Y: Hoehn and Yahr scale; MoCA: Montreal Cognitive Assessment;
ND: no data]
+ 5 data points missing in apathy, Beck depression scale, and MoCA
3.2.2 Image acquisition
Images of both cohorts were acquired with a Philips Achieva 3.0 Tesla MRI scanner (Best, The
Netherlands). 3D T1-weighted images were acquired with CLEAR homogeneity correction with
1×1×1 mm3 resolution. Eight minutes of resting-state functional MRI (rsfMRI) data were acquired
with an echo-planar imaging sequence with 3×3×3 mm3 resolution, 36 slices, 2000 ms TR, 30 ms
TE, and 240 dynamics.
3.2.3 Image preprocessing
Several image preprocessing steps were applied to the fMRI data including slice-timing, isotropic
reslicing, and motion correction in MATLAB using SPM8 functions (the Wellcome Trust Centre
68
for Neuroimaging, UK) and in-house Matlab codes. FLIRT (the FMRIB Centre, UK) was used to
register fMRI images and structural images. T1-weighted images were used for cortical and
subcortical parcellation, which was carried out using FreeSurfer software 4.5.0 (Massachusetts
General Hospital, USA). For MS subjects, thirty-eight cognition-related regions-of-interest
(ROIs), which were commonly reported in the MS neuropsychology literature, were chosen (Table
3.3) for connectivity analysis. For PD subjects, fifty-four ROIs that were clinically relevant for PD
were chosen (Table 3.4). As the original purpose for the MS project was to evaluate functional
connectivity and its relation to cognition, included ROIs were selected based on the
neuropsychology literature so these ROIs were supposed to be highly associated with cognition.
However, the original purpose for this PD project was to investigate the connectivity among the
most clinically affected regions in PD regardless of motor or cognitive syndromes. Therefore, the
selected ROIs were not the same, which is one of the limitations in this chapter. The mean time
courses over all voxels within one region in fMRI data were extracted from the given ROIs. All
calculations were done in the subject’s native space. As the processing pipeline was original
designed based on common preprocessing steps in SPM package, further steps such as temporal
filtering and nuisance signal regression were not applied. However, linear detrending was applied
on the extracted time courses and voxel outliers were removed. Typically, we exclude subjects
who show higher than 2 mm in translation and/or 2 degrees in rotation, but none of the subjects
reached these criteria.
Bilateral ROIs
frontal pole parietal and occipital junction areas
superior frontal gyrus superior occipital gyrus
middle frontal gyrus anterior cingulate cortex
69
inferior prefrontal cortex posterior cingulate cortex
temporal pole, insula cortex, amygdala precuneus
superior temporal cortex medial orbitofrontal cortex
posterior parietal cortex lateral orbitofrontal cortex
post central gyrus fusiform gyrus
supramarginal gyrus superior parietal cortex
medial temporal lobe, hippocampus, parahippocampal gyrus
Table 3.3 The 38 ROIs in the study for MS (19 bilateral regions).
Bilateral ROIs
middle frontal gyrus lateral occipital region
ventral lateral prefrontal gyrus caudal anterior cingulate cortex
insula posterior cingulate cortex
superior temporal gyrus precuneus
middle temporal gyrus ventral medial prefrontal cortex
inferior temporal gyrus cerebellum
parahippocampal gyrus primary motor cortex
hippocampus supplementary motor area
superior parietal gyrus pre supplementary motor area
inferior parietal lobe premotor area dorsal part
occipital-parietal area premotor area ventral part
thalamus putamen
caudate pallidum
Table 3.4 The 54 ROIs in the study for PD (27 bilateral regions).
3.2.4 Functional connectivity analysis
For assessing interhemispheric connectivity, we computed correlation coefficients using a
combination of two approaches. For each subject, we first divided the time courses from n ROIs
70
into n/2 homologous left, right pairs. For the ith pair, Pi(L,R), we examined the remainder right and
left homologous pairs P1…n/2,≠i (L,R) and computed both the simple correlation (Pearson’s r) and
the partial correlation conditioned on Pi(L,R) (i.e. the rest ROIs). We then computed the difference
in the sum of the correlations and partial correlations between each homologous pair, Pi(L,R). The
calculation could be expressed as the following mathematical style pseudocode:
∑ ( | (partial correlation values – simple correlation values) | ).
In order to determine if interhemispheric connectivity was associated with altered performance on
cognitive tests, we performed robust regression, which is a common method to predict clinical or
behavioural responses using neuroimaging data [Bowman, 2014a]. We took all homologous pair
connectivity difference values from every subject as independent variables and behavioural scores
as the dependent variable. Age was also included as a nuisance covariate. In MS, the behavioural
scores included SDMT, PASAT, EDSS and LCVA scores. In PD, due to small sample size, only
the significant connectivity pairs were used for regression analysis as robust regression cannot take
more features (connectivity pairs) than observations (sample size). All the clinical scores were
included except MoCA, apathy, and Beck depression scale because missing data points resulted in
insufficient data to perform robust regression. The first calculation with robust regression obtained
β values, which represented the weighting between interhemispheric connectivity and behavioural
scores. These values were then included in the second calculation to generate predicted
cognitive/clinical scores. Finally, R-squared between raw and predicted scores was reported.
Figure 3.1 explains the procedure in a graph fashion.
71
Figure 3.1 The procedure of regression analysis. The purpose of the 1st calculation (the left panel) is to obtain
β values (highlighted with yellow) in a robust regression model. In the 2nd calculation (the right panel), these
β values are included in a linear regression to calculate predicted scores (highlighted in red).
In order to determine if certain homologous pairs had the largest influence on predicting clinical
and cognitive scores, we repeated the regression with a sparsity penalty on the regression
coefficients by implementing elastic net regression (lasso function in Matlab, with alpha = 0.75)
and included age as a covariate. This algorithm attempts to find a balance between predicting
cognitive scores and including as few interhemispheric connections as possible to make that
prediction. We used a leave-one-out cross validation to determine the optimal shrinkage parameter
λ.
For evaluating whether long-range connections were different between patients and HC, we
applied an exploratory approach by calculating the whole brain connectivity matrix using
Pearson’s r correlation. The matrix elements were then transformed to Fisher’s z values and then
converted into a vector for the two-sample t-test with false discovery rate (FDR) corrected. The
vectors were reshaped back to matrices and significant differences of connectivity pairs between
groups were reported.
72
3.3 Results
Conditioning on the different left ↔ right ROI pairs resulted in significant differences between
partial and simple correlation in MS (Figure 3.2). The changes in overall instantaneous
interhemispheric connectivity were significantly different between MS and HC (corrected p =
0.05) when the following ROI left ↔ right pairs were used for the partial correlation: the superior
parietal cortex, superior occipital gyrus, and precuneus. In addition, the MS group constantly
showed higher differences than HC group no matter what pairs were conditioned. In PD,
significantly different interhemispheric connectivity was only observed between HC and PD on-
medication when the ventral medial prefrontal cortices was used for partial correlation (corrected
p = 0.05). Similar to MS, PD subjects also showed higher differences than HC regardless which
pairs were conditioned (Figure 3.3). The differences represent the relative importance
(connectivity values) of the brain regions (conditional ROI) when it is included in the analysis. A
larger difference reflects greater importance of that region.
73
Figure 3.2 Functional interhemispheric connectivity in normal controls and MS. The overall pattern
demonstrates that in MS, the differences are higher than in normal subjects across all regions. This suggests
greater overall interhemispheric connectivity in MS.
[front-pole: frontal pole; front-sup: superior frontal gyrus; front-middle: middle frontal gyrus; inf-
prefrontal: inferior prefrontal cortex; tem-pole/Ins/Amyg: temporal pole and insula and amygdala merged;
sup-temp: superior temporal cortex; med-temp/hip/parahip: midial temporal lobe and hippocampus and
parahippocampus merged; postcentral: post central gyrus; sup-par: superior parietal cortex; suprama:
supramarginal gyrus; par-occi: parietal and occipital junction areas; sup-occi: superior occipital gyrus;
cACC: caudal anterior cingulate cortex; PCC: posterior cingulate cortex; precun: precuneus; med-OFC:
medial orbitofrontal cortex; lat-OFC: lateral orbitofrontal cortex; fusiform: fusiform gyrus; pos-par:
posterior parietal cortex]
74
Figure 3.3 Functional interhemispheric connectivity in healthy controls, PD on-medication, and PD off-
medication. The only difference is found between HC and PD on-medication. The overall pattern
demonstrates that PD shows higher differences than HC across all regions. This suggests greater overall
interhemispheric connectivity in PD, especially in off-medication state.
[M1: primary motor cortex; SMA: supplementary motor area; PMd: premotor area dorsal part; PMv:
premotor area ventral part]
Using robust linear regression, SDMT and PASAT scores could be accurately predicted from the
differences between interhemispheric partial and simple correlation (Figure 3.4); while EDSS and
LCVA results were insignificant. We performed elastic net regression to detect if only a few of
the interhemispheric connections were responsible for prediction (Figure 3.5). However, even with
the sparsity constraint (as we set α = 0.75, the elastic net regression model was more toward L1
norm of β (lasso regression) than L2 norm of β (ridge regression), which sparsity constraint was
taken into account.) on the regression coefficients, 18 and 10 ROI pairs respectively were still
75
included in the model to predict SDMT and PASAT scores. This method is able to show which
linear combination of ROI pairs plus age best predict cognitive performance (i.e. linear
combination of positive plus negative weightings). The frontal pole, temporal pole, insula,
amygdala regions, superior temporal gyrus, parietal and occipital junction areas, anterior cingulate,
and posterior parietal cortex were found to have more positive weights on SDMT performance. In
contrast, the superior temporal gyrus, medial temporal gyrus, hippocampal regions, supramarginal
areas, superior occipital gyrus, and anterior cingulate cortex impacted PASAT more than other
ROI pairs.
Figure 3.4 Relationships between real scores and predicted scores in SDMT and PASAT tests based on cross-
validation. The predicted scores are calculated based on interhemispheric connectivity values and ß values
in the linear regression model. Raw and predicted SDMT scores demonstrate a strong correlation (R2=0.97)
as well as raw and predicted PASAT scores (R2=0.83). The 45-degree lines indicate perfect predictability.
[PASAT: Paced Auditory Serial Addition Test; SDMT: Symbol Digit Modalities Test]
76
Figure 3.5 Important brain regions for cognitive performances in MS. With elastic net regression, the upper
left and upper right panels express weights of important connectivity pairs in predicting SDMT and PASAT,
respectively. The lower panel shows positive correlations between real scores and predicted scores of SDMT
and PASAT with the limited number of pairs. Real and predicted SDMT scores especially demonstrate a
good correlation with R2 = 0.96. The 45-degree thick-red lines indicate perfect predictability. The thin-red
lines indicate linear fitting. The two lines are overlapped in SDMT.
77
[front_pole: frontal pole; front_sup: superior frontal gyrus; front_middle: middle frontal gyrus;
inf_prefrontal: inferior prefrontal cortex; tem_pole/Ins/Amyg: temporal pole and insula and amygdala
merged; sup_temp: superior temporal cortex; med_temp/hip/parahip: midial temporal lobe and
hippocampus and parahippocampus merged; postcentral: post central gyrus; sup_par: superior parietal
cortex; suprama: supramarginal gyrus; par_occi: parietal and occipital junction areas; sup_occi: superior
occipital gyrus; cACC: caudal anterior cingulate cortex; PCC: posterior cingulate cortex; precun:
precuneus; med_OFC: medial orbitofrontal cortex; lat_OFC: lateral orbitofrontal cortex; pos_par:
posterior parietal cortex]
In PD, due to insufficient data, we cannot perform elastic net regression and only the significant
connectivity pair in the t-test (i.e. ventral medial prefrontal connectivity) was used in robust
regression. Only medication dose was predicted with R-square 0.2 in PD on-medication. The other
scores were all predicted poorly with R-square lower than 0.2 in both on and off medication status
(results are shown in Appendix B.1).
Figure 3.6 shows the connectivity pairs that are significantly different between MS and HC in the
connectivity matrices (p<0.05, FDR corrected). Overall, most of the significant connections linked
frontal and parietal/occipital regions as well as frontal and temporal areas, which can be seen as
long-range connections. The connections that link two hemispheres were also seen. Figure 3.7
demonstrates the significant connections which distinguish PD on-medication and PD off-
medication (p<0.05, FDR corrected). These connections linked the caudate, premotor area, and
middle frontal gyrus, which is part of the frontostriatal loops and the output component of the
executive function model [Miller and Cohen, 2001] (see section 1.4.2). No significant connections
were found between HC and PD regardless patients’ medication status.
78
Figure 3.6 Connections that are significantly different between HC and MS. In addition to the connections
which link two hemispheres (not restricted to homologous regions), many connections link frontal and
parietal/occipital regions as well as frontal and temporal areas distinguish two groups. In total, there are 163
connections. The figure is created using BrainNet Viewer [Xia et al., 2013].
Figure 3.7 Connections that are significantly different between PD off-medication and
PD on-medication. These connections link the middle frontal gyrus, caudate, and
ventral premotor area. The figure is created using BrainNet Viewer [Xia et al., 2013].
79
3.4 Discussion
3.4.1 Enhanced interhemispheric connectivity may indicate compensatory mechanisms
Previous fMRI studies have reported decreased functional connectivity in MS [Dogonowski et al.,
2013; Kingwell, 2012]; while increased functional connectivity has also been reported during rest
as well as during performance of tasks [Janssen et al., 2013; Kingwell, 2012; Specogna et al.,
2012], which may reflect compensatory mechanisms. For example, within the default mode
network areas, several cortices demonstrate enhanced functional coupling in MS, which is
associated with a concomitant loss of cognitive efficiency [Hawellek et al., 2011a]. Consistent
with prior studies, we found overall enhanced interhemispheric connectivity in MS subjects as
demonstrated in Figure 3.2 where the MS group constantly showed bigger differences. The
differences measured alterations between connectivity values derived from two methods (simple
correlation and partial correlation). Hence, bigger differences also indicated higher connectivity.
In this study, we discovered significantly altered functional interhemispheric connectivity in MS
through an approach which specifically assessed homologous connections.
In PD, previous studies which specifically measured connectivity between homologous regions all
reported decreased functional interhemispheric connectivity [Luo et al., 2015b; Zhu et al., 2016].
However, the approach (i.e. VMHC) utilized in these studies is very different from our approach.
Simply calculating Pearson’s r correlation between homotopic voxels may capture effects from
other regions and voxels as simple correlation does not measure “direct correlation”. Moreover,
the disease severity in our cohort is different from those studies. If we interpret increased
interhemispheric connectivity in this study as a compensatory effect, the disease severity of the
study cohort plays an important role and it could be one of the reasons for distinct findings. A
80
model of network collapse has been proposed in MS, which describes a non-linear relation between
functional connectivity, cognitive impairments, and structural damage across disease course
[Schoonheim et al., 2015]. Although MS and PD have unique pathologies, we believe that there
are certain similarities between them as discussed in Chapter 1 & 2. Therefore, such model may
apply to PD as well. A study which recruited PD patients with similar disease severity (based on
H & Y scale) also demonstrated increased rsFC as a compensatory mechanism [Simioni et al.,
2016]. In addition, this study also observed that patients in an off-medication state showed
increased connectivity in the cerebellum, primary motor cortex, and subcortical regions; while in
on-medication state, these stronger connections were reduced and both on/off-medication rsFC
were correlated with/predicted motor performance, confirming that such increased rsFC pattern
reflected a compensatory mechanism. Our results also support that increased rsFC in PD may
represent compensation. Similarly to MS, PD also demonstrated higher interhemispheric
connectivity than HC especially in off-medication state (Figure 3.3). In fact, not only
interhemispheric connections, but the whole brain connectivity in off-medication exhibited higher
rsFC than HC (Appendices B.2). Although we did not perform regression analysis to probe
whether cognitive performance was related to such compensatory pattern (only medication dose
can be roughly predicted, Appendices B.1), our findings still provide insights into the disease
affects neuronal networks in PD.
We used both partial and simple correlation to investigate the interhemispheric connectivity. This
approach has not been used in the previous research assessing interhemispheric connectivity,
which makes the methodology in this study unique. Given that the calculation estimated
correlation differences only between homologous brain regions with and without conditioning, it
81
is more straightforward to interpret the results as interhemispheric. Since the correlation
differences were robustly larger in MS and PD (Figure 3.2 & 3.3) and the overall interhemispheric
connectivity was also larger in the patient groups, there is both enhancement and homogenisation
of interhemispheric connectivity in neurological disorders, meaning that the interhemispheric
connections are more dependent on each other. If one connection is altered, the rest of the
connections are more likely to be influenced as well. This also implies that interhemispheric
connections in MS and PD lose their ability to independently modulate connectivity. Instead, the
homologous regions require further support from other connectivity pairs in order to transfer
information across the two hemispheres, which indicates that enhanced interhemispheric
communication could be an early compensatory change in MS and PD.
3.4.2 Interhemispheric connectivity predicts cognitive performance
In addition to the finding of overall interhemispheric connectivity, we have also demonstrated that
altered functional interhemispheric connectivity closely reflects performance on cognitive tests,
namely the SDMT and PASAT in MS (Figures 3.4 & 3.5). Interestingly, even when we used a
sparsity constraint on the regression in order to see if only a few interhemispheric connections
were important for SDMT and PASAT performance, we still found several ROI pairs were
required for accurate prediction of cognitive performance. Thus both SDMT and PASAT
performance appear to rely on widespread interhemispheric connectivity. Nevertheless, some
cortical regions, which showed positive and high weightings with SDMT and PASAT, had greater
influence on predicting cognitive scores such as the insula, frontal areas, temporal areas, anterior
cingulate cortex, supramarginal areas, and parietal/occipital regions. Many of these areas are
considered “hub” regions in distributed brain networks and reflect critical waypoints for
82
information transfer [Achard et al., 2006; Hagmann et al., 2008; van den Heuvel and Sporns,
2011], playing an important role in maintaining connectivity and information integration (a key
aspect of cognition) in anatomically segregated brain networks. Since hub regions underline
numerous aspects of cognition, we believe that these interhemispheric connectivity pairs
dramatically influence the cognitive processes which are involved in SDMT and PASAT
performance. In addition, it has been suggested that hub regions may be particularly sensitive to
functional disruption [Hagmann et al., 2008; van den Heuvel and Sporns, 2011], which is perhaps
another reason why both SDMT and PASAT are sensitive measures of cognitive dysfunction in
MS. Among the clinical tests, LCVA did not show significant association with interhemispheric
connectivity compared to other tests, suggesting that vision problems are less likely to impact
interhemispheric connections in our cohort. The results further emphasize the importance of
interhemispheric connectivity in cognition.
In PD, unfortunately, due to sample size, we did not have enough results to probe whether
interhemispheric connectivity could accurately predict cognitive function, however, we do not
necessarily exclude this possibility. In fact, studies which evaluate structural interhemispheric
connectivity have shown associations between worsening cognitive function and reduced white
matter integrity of interhemispheric fiber tracts in parkinsonism [Fling et al., 2016]. In an MEG
study, cognitive symptoms were related to increased interhemispheric connectivity of alpha
frequency estimated by synchronization likelihood in newly-diagnosed PD patients [Stoffers et al.,
2008]. Taken together, both structural and functional interhemispheric connectivity show relations
with cognitive functions in PD and related disorders to certain degree; however, more research on
the effects of interhemispheric connections to cognition in PD is needed to draw a firm conclusion.
83
3.4.3 Sensitivity of interhemispheric connectivity to different cognitive tests
Although both PASAT and SDMT are clinically used for investigating cognitive deficits in MS,
numerous studies have reported the SDMT to be a more sensitive, valid and reliable measure than
the PASAT [Drake et al., 2010; Langdon et al., 2012; Parmenter et al., 2007; Sonder et al., 2014;
Strober et al., 2009]. Indeed, we found a very strong correlation between SDMT and
interhemispheric connectivity measures (Figure 3.4), consistent with prior studies suggesting
SDMT is more strongly associated with MRI measures [Benedict et al., 2006]. With the elastic net
regression, we show that connectivity is important between frontal pole, temporal pole, insula,
amygdala, superior temporal cortex, parietal and occipital regions, and posterior parietal cortex for
SDMT performance (Figure 3.5), which are partially consistent with previous studies showing that
parietal areas play a role in SDMT as well as frontal areas and occipital regions [Forn et al., 2011].
The other regions that we found, such as temporal pole, insula, amygdala, and superior temporal
cortex have not been mentioned in previous studies. We suspect that these regions act as mediators
in information communication between frontal and parietal areas for a short time, which cannot be
observed by the methods used in previous studies. In contrast, performance on PASAT may require
information coordination between frontal, parietal regions, and the cerebellum since activation
patterns mainly locate within one hemisphere among these regions [Forn et al., 2011], which will
not be captured by the interhemispheric connectivity measures examined here but shall be
encapsulated in long-range connectivity.
3.4.4 Altered long-range connections
As Chapter 1 hypothesized, cortico-to-cortical long-range connections, which are important for
executive function, are impaired in neurological disorders. We have shown that the connections
84
which significantly distinguish MS and HC were cross hemispheric connections and long-range
connections primarily connecting frontal, temporal, parietal, and occipital regions as shown in
Figure 3.6. Interestingly, such connectivity pattern was correlated with PSAST but not SDMT
performance (Appendices B.3), supporting the idea in the section 3.4.3 that PASAT performance
may require coordination between frontal, parietal, cerebellar regions (i.e. long-range connections)
more than SDMT performance. On the other hand, in PD, connections which primarily linked the
middle frontal gyrus and caudate were significantly different between PD on and off medication
states. This connectivity pattern is partially consistent with the output component mentioned in
Chapter 1 and the frontostriatal circuits in the PD literature. Due to the fact that the PD subjects
we assessed were mild and not cognitively impaired (average MoCA>26), perhaps long-range
connections between frontal, parietal, and occipital regions were less impaired compared to the
MS cohort. In addition, given that PD patients demonstrate cognitive inflexibility and such deficit
has been linked to neuronal activity [Cools et al., 2001; Lange et al., 2017], rsFC might be less
flexible as well, which cannot be captured in the current analysis. Instead, time-varying approaches
are able to detect temporal changes of rsFC and estimate dynamic functional connectivity. This
aspect in PD will be discussed in a later chapter.
3.5 Limitation
We emphasized ROIs associated with higher order cognition based on the previous clinical
literature, and thus would not detect other types of connectivity disruption affecting lower
cognitive domains such as vision. Moreover, as mentioned in section 3.2, the original study goals
of two projects were different so the selected ROIs were not the same, which causes some difficulty
in directly comparing results between two disease populations. In a future study, ROI selection
85
could be more consistent. In addition, here we aimed to study cortical connectivity so some
subcortical regions, which might be important for high order cognition, were not included.
Therefore, our data were not sufficient to investigate cortico-striatal loops in MS for example. Due
to the study design, we did not administer a comprehensive neuropsychological test battery to
examine all cognitive domains. Therefore, the results only represent the cognitive tests which have
been commonly used in clinical trials and clinical screening. Nonetheless, the PASAT and SDMT
are well validated, sensitive and widely used measures of cognition in neurological disorders and
inclusion of these measures allows us to compare to previous imaging and cognition studies. Future
studies should seek to determine interhemispheric connectivity across a wide range of cognitive
domains. While we made attempts to prevent overfitting, including cross-validation, more subjects
would enhance the robustness of our results. In addition, the patients were not recruited based on
cognitive impairment so our results might not be representative of neurological disease cohorts
with severe cognitive problems. Finally, fMRI feature selection for assessing relations between
rsFC and cognitive functions remains challenging. Including too few features may lead to
insufficient information for brain-behaviour analysis, but having too many features often causes
overfitting in neuroimaging research. How to extract informative rsFC characteristics has become
an important need. Several methods have been proposed to summarize rsFC patterns such as graph
theoretical analysis and other advanced network approaches [Fornito et al., 2013; Rubinov and
Sporns, 2010]. These approaches will be discussed in the following chapters.
3.6 Conclusion
This study emphasizes the importance of intact functional interhemispheric connectivity in MS
and PD, particularly with respect to cognitive performance. We have demonstrated the
86
characteristics of interhemispheric connectivity in neurological conditions and healthy controls:
these connections are enhanced and become more homogeneous in patients. Furthermore, in MS,
SDMT and PASAT scores were robustly correlated with interhemispheric connectivity.
Exploratory analysis revealed that long-range connections were altered in MS and such
connectivity pattern was related to PASAT performance; while the connections between the frontal
and caudate regions distinguished PD on and off medication states but were not related to any
clinical scores. The results may potentially benefit the development of novel treatments for
cognitive deficits in neurological disorders: targeting interhemispheric connections (as well as
long-range connections) and providing cognitive training may help with maintaining normal
cognitive functioning.
87
Chapter 4: Brain organization and executive function
In this chapter, graph theoretical analysis is carried out to investigate the brain organization in PD
and MS, which includes functional integration, functional segregation, and hub structures.
Moreover, correlation and regression analyses are implemented to explore the relations between
brain organization, cognitive performances, and disease severity in both neurological populations.
A simple analysis is conducted to test the reproducibility of these graphical measures.
4.1 Introduction
4.1.1 Connections and networks of rsFC and the relations to cognition
Many neuroimaging studies attempt to link cognitive deficits and functional connectivity (FC) at
rest. Impaired corticostriatal connectivity resulting in decreased integration among the striatum,
mesolimbic cortex, and sensorimotor cortex has been associated with some non-motor symptoms
in PD such as mental “rigidity” [Luo et al., 2014]. Overall global cognitive performance in PD is
shown to be associated with decreased FC in widespread regions including the paracentral lobe,
superior parietal lobe, occipital regions, inferior frontal gyrus, and superior temporal gyrus [Olde
et al., 2014]. Weakened FC in the frontoparietal network has been shown to be related to worsening
executive function in PD with mild cognitive impairment (MCI) [Amboni et al., 2014a]. Therefore,
it is speculated that not only the frontostriatal connections but whole-brain altered connectivity
contributes to cognitive deficits in PD.
88
In MS, decreased connectivity indicates that cortical regions fail to integrate information in
resting-state networks (RSNs) derived from independent component analysis (ICA), especially the
medial prefrontal cortex and posterior cingulate cortex in the default mode network (DMN), and
has been associated with worsening executive function in cognitively impaired patients [Cruz-
Gómez et al., 2014; Louapre et al., 2014]; however, another study which calculated whole brain
Pearson’s r matrices reported that increased functional coupling of rsFC within the DMN regions
was related to worsening cognitive efficiency, which was measured by several tests reflecting
executive function [Hawellek et al., 2011b]. With a seed-based approach, increased functional
connectivity between “cognitive hubs” and the cerebellum, middle temporal gyrus, occipital pole,
and angular gyrus was associated with better performance of executive ability, but this study
focused on limited brain regions and may neglect ROIs which are not mentioned in the traditional
psychology literature [Loitfelder et al., 2012].
4.1.2 Graphical measures and cognition
One of the strategies to characterize the whole brain FC is to apply graph theory analysis to
summarize the overall network of connections between Regions of Interest (ROIs). These network
characteristics summarize the whole brain FC globally and locally in terms of functional
integration, segregation, and core/hub structures [Rubinov and Sporns, 2010; Sporns, 2013].
Several network measures have been linked to human intelligence and cognitive functions in
healthy subjects [Cohen and D’Esposito, 2016; Pamplona et al., 2015]. In PD, regions in the
orbitofrontal regions and occipital pole show decreased node degree (i.e. fewer numbers of
connections to/from a given region); while superior parietal, posterior cingulate cortex,
supramarginal, and supplementary motor areas have increased node degree [Göttlich et al., 2013].
89
Baggio et al. [Baggio et al., 2014] discovered reduced FC in long-range connections (i.e.
connections between frontal, occipital, and parietal areas) in both PD and PD with MCI; while
increased graph theoretical measures such as clustering coefficient, local efficiency, and
modularity in the frontal areas were negatively correlated with attention/executive scores in PD
with MCI, possibly as compensation/adaptation for impairments in long-range connections.
Patients with MS exhibit reduced global efficiency, node degree, centrality, and increased path
length on average compared to normal controls, demonstrating altered rsFC in network
organization as well as implying inefficient information transfer in the brain in MS [Rocca et al.,
2016a]. Another study found that only local measures (local efficiency and clustering coefficient
in the left insula, inferior frontal areas, and cuneus) were decreased in MS, but these alterations
did not show any associations with cognitive scores in a correlation analysis [Shu et al., 2016]. In
addition, an article concluded that increased modularity of rsFC in early stage MS, which
represented diminished functional integration, was correlated with worse performance in a dual
task [Gamboa et al., 2014]. Overall, although there are minor discrepancies between studies,
research with graph theory implies that rsFC in MS is more segregated and may be less efficient
to support complex cognitive tasks.
4.1.3 Hub structures and cognition
Previous work has suggested that “hub regions” may be particularly vulnerable to
neurodegenerative processes [Crossley et al., 2014]. In healthy subjects, the insula and several
frontal regions such as the superior frontal gyrus act as hubs and may facilitate cognitive processes
to an even greater extent than the caudate [Baggio et al., 2015; van den Heuvel and Sporns, 2013].
In PD, there is a tendency of failure in hub regions and non-hub regions become more important
90
than that in healthy subjects [Koshimori et al., 2016]. Reduced importance of normal hubs in the
frontal areas and increased importance of non-hub regions in other cortices has been observed in
PD with MCI [Baggio et al., 2014]. In PD, the insula may be less likely to act as a hub, but there
may be an enhanced role of the caudate as a hub [Koshimori et al., 2016]. This altered brain
organization may indicate that the disease first impairs hubs that are important for cognition,
resulting in cognitive dysfunction in PD.
On the other hand, in MS, decreased importance in sensorimotor and ventral stream regions has
been related to higher disease severity and poor cognition, respectively [Schoonheim et al., 2013].
In MS with MCI, increased connectivity in the DMN and frontoparietal network, which engage
many hub regions, is correlated with worse cognitive performance, stating that cognitive
impairment affects distributed communication between non-hub regions in “hub-rich” networks
[Meijer et al., 2017].
4.1.4 Study aims
In this study, we aimed to investigate 1) functional connectivity changes in PD and MS subjects
measured by graph theory analysis, and 2) how cognition is associated with altered graphical
measures in PD and MS. Instead of only investigating FC in specific circuits such as the
frontostriatal loops, we studied whole-brain connectivity and characterized it by graph theory
measures. Finally, we utilized correlation analyses to explore the associations between FC and
cognitive profiles and performed regression analyses to predict behavioural outcomes.
Advanced network analyses have been increasingly conducted and applied to clinical
neuroimaging research. As there is a growing trend to include these measures in rsfMRI studies,
91
reproducibility of such features needs to be addressed. Some studies have shown that several
graphical measures demonstrate high reliability and reproducibility such as global efficiency, path
length, and clustering coefficient among others [Braun et al., 2012; Shah et al., 2016; Telesford et
al., 2010]. We also conducted a simple analysis to test whether these graphical measures were
reproducible across scans.
4.2 Materials and methods
4.2.1 Subjects and behavioural data
Eleven healthy subjects (mean age: 25.7±3.6, 4 females, 7 males, all with university education
level) were recruited to test the reproducibility of advanced network measures. None of these
subjects showed neurological or psychiatric conditions.
The PD data are from the Parkinson's Progression Marker Initiative (PPMI) and MS data are from
the COGMS project -- the same as chapter 2. Only a subset of MS subjects were used to ensure
that the healthy subjects used for comparison were age-matched healthy subjects.
Thirty-one PD patients who enrolled in the PPMI were included in this study. Detailed descriptions
are provided in section 2.2.1. For imaging analysis, twenty-three subjects were included as 8
subjects had excessive imaging and motion artifacts and they were removed from the analysis.
Only one subject in the final cohort showed MCI. For comparison, nineteen age-matched healthy
subjects (HS) were recruited in this study through Pacific Parkinson’s Research Centre at UBC
Hospital. None of the HS showed cognitive impairment screened by MoCA. Ethics approval was
issued by the University of British Columbia's Research Ethics Board and all subjects provided
92
written, informed consent. Table 4.1 shows the demographical characteristics of PD and HS
subjects.
Forty-six Relapsing-Remitting Multiple Sclerosis (RRMS) patients were included in the study and
all the subjects underwent both cognitive testing and MRI scanning. Demographics are shown in
Table 4.2. A subset of MS subjects and age-matched normal controls (NC) with university
education were included for comparison purposes (18 MS and 15 NC, mean age±SD in MS/NC:
32.00±4.93/28.93±5.00) (Table 4.2). All patients fulfilled the McDonald 2005 criteria [Polman et
al., 2005] for the diagnosis of MS and were recruited from the MS clinic at the University of British
Columbia Hospital. Exclusion criteria included: 1) subjects with significant depression and/or
other psychiatric illness, 2) history of drug or alcohol abuse, or 3) use of steroids in the last 3
months.
After the exclusions for movements and imaging artifacts, the final PD cohort included 23 subjects
and 19 healthy controls. The data were used in both measure comparison and brain-behaviour
analysis. The final MS cohort included 46 MS subjects for brain-behaviour analysis. For measure
comparison, a subset of the MS subjects were chosen (18 subjects) and 15 healthy controls were
included in order to be age-matched.
Parkinson’s subject (mean±SD) healthy subject (mean±SD)
demographics & clinical data gender 10 females/13 males 9 females/10 males
age b 61.04±9.8 56.12±16.9
UPDRS b 16.96±11.2 no data
depression b 5.17±1.1 no data
education in years b 17.17±2.8 no data
cognitive scores
MOCA b 27.48±2.2 27.39±1.6a
BJLOTOT b 25.30±4.2 no data
93
HVLTTOT b 23.09±5.8 no data
HVLTDELAY b 8.39±2.7 no data
DVT-HVLTTOTAL b 44.70±15.4 no data
DVT-HVLTDELAY b 49.00±16.0 no data
DVT-HVLTRETENTION b 49.61±13.0 no data
LNS-RAW b 10.74±2.6 no data
SFCOM b 48.22±10.6 no data
SFVEG 13.74±3.9 no data
SFANI 21.52±4.9 no data
SFFRU 12.96±4.1 no data
SDMT b 41.57±9.3 no data
Table 4.1 Demographics, clinical data, and cognitive scores in PD and HS.
[UPDRS = Unified Parkinson's Disease Rating Scale, MOCA = Montreal Cognitive Assessment, BJLOTOT =
Bento Line Orientation Total Score, HVLTTOT = Hopkins Verbal Learning Test-Revised Total Score,
HVLTDELAY = HVLT Delayed Recall Score, DVT-HVLTTOTAL = standardized HVLT Total Score, DVT-
HVLTDELAY = standardized HVLT Delayed Recall Score, DVT-HVLTRETENTION = standardized HVLT
Recognition Trial Score, LNS-RAW = raw Letter-Number Sequencing Test Score, SFCOM = Sematic Fluency
Test – combination, SFVEG = Sematic Fluency Test – vegetable trial, SFANI = Sematic Fluency Test – animal
trial, SFFRU = Sematic Fluency Test – fruit trial, SDMT = Symbol Digit Modalities Test]
a: one data point is missing
b: the tests that were included in brain-behaviour analysis
46 MS subjects
(mean±SD)
18 MS subjects
(mean±SD)
15 NC subjects
(mean±SD) Age 42.89±10.9 32.00±4.9 28.93±5.0
Gender 13 males/33 females 3 males/15 females 7 males/8 females
EDSS 2.47±1.8 1.63±1.5 no data
Education (years) 14.80±2.5 15.67±2.5 18a
Disease Duration
(years)
11.45±8.7 5.46±3.8 no data
Table 4.2 Demographics in MS and NC.
a Due to many missing data points, 18 is an estimated number.
94
The neuropsychological assessments in PD and MS have been described in section 2.2.2. Detailed
cognitive scores of 46 MS subjects are shown in Table 2.2. Overall, attention, executive function,
visuospatial ability, and memory domain were evaluated in PD and 14 behavioural scores were
included in brain-behaviour analysis as indicated in Table 4.1. Attention, executive function,
processing speed, and working memory domains were assessed in MS and 14 behavioural scores
were included in brain behaviour analysis such as age, education, EDSS, disease duration,
Working Memory Index (WMI), Processing Speed Index (PSI), Verbal Fluency Test (FAS),
Wisconsin Card Sorting Test Complete Categories (WCST-CC), transformed Trail-Making-Test
A and B (tTMTA/B), Multiscale Depression Inventory (MDI), State-Trait Anxiety Inventory State
(STAIS), State-Trait Anxiety Inventory Trait (STAIT), and Fatigue Severity Scale (FSS).
4.2.2 Imaging acquisition
For testing the reproducibility, three resting state fMRI (rsfMRI) sessions were scanned
continuously on a Philips Achieva 3.0 Tesla MRI scanner with an echo-planar imaging (EPI)
sequence with the following parameters: 3×3×3 mm3 resolution, 36 slices, 2000 ms Repetition
Time (TR), 30 ms Echo Time (TE), 90 degrees flip angle, and 240 volumes/dynamics (8 minutes
in total). 3 Dimensional (3D) T1 weighted images were acquired with 1×1×1 mm3 resolution, 60
slices, 28 ms TR, 4 ms TE and 27 degrees flip angle.
In PD subjects, a standardized MRI protocol on a 3 Tesla Siemens Trio Tim MR system was used.
3D T1-weighted structural images were acquired using MPRAGE GRAPPA sequence with TR
2300 ms, TE 2.98 ms, Field of View (FoV) 256 mm, and resolution 1x1x1 mm3. RsfMRI images
were acquired using echo-planar imaging (EPI) to detect Blood Oxygenation Level Dependent
(BOLD) contrast with 212 volumes, 40 slices in ascending direction, TR 2400 ms, TE 25 ms, FoV
95
222 mm, and resolution 3.3×3.3×3.3 mm3. All HS subjects underwent imaging scans in MRI
research centre at UBC Hospital with a Philips Achieva 3.0 Tesla MRI scanner. 3D T1-weighted
images were acquired with TR 8 ms, TE 4 ms, FoV 256 mm, and resolution 1×1×1 mm3. RsfMRI
data were acquired with EPI sequence and the following parameters: 186 volumes/dynamics, 36
slices in interleaved direction, TR 2500 ms, TE 30 ms, FoV 240 mm, and resolution 3×3×3.97
mm3.
Both MS and NC subjects underwent imaging studies at the University of British Columbia (UBC)
MRI Research Centre. Resting-state functional MRI (rsfMRI) data were acquired using an 8
channel head coil and an EPI sequence with the following parameters: 3×3×3 mm3 resolution, 36
slices, 2000 ms TR, 30 ms TE, 90 degree flip angle, and 240 volumes/dynamics. 3D T1 weighted
images were acquired with 1×1×1 mm3 resolution, 60 slices, 28 ms TR, 4 ms TE and 27 degree
flip angle.
During rsfMRI scan, all subjects were instructed to rest quietly with their eyes closed and not to
fall asleep.
4.2.3 Preprocessing
The preprocessing steps were the same as chapter 3. Image preprocessing steps were performed in
each subject’s native space with the functions of slice timing and motion correction from Statistical
Parametric Mapping 8 (SPM8, University College London, London) for correcting temporal and
spatial differences. For registration, the FMRIB's Linear Image Registration Tool (FLIRT) from
the FMRIB Software Library 6.0 (FSL, FMRIB, Oxford) was used and a brain mask was applied
to remove non-brain areas before registration. Moreover, fMRI images were rescaled to isotropic
96
using a self-programmed script in Matlab (The MathWorks, Inc.), so all fMRI data were 3×3×3
(MS and all healthy subjects) and 3.3×3.3×3.3 (PD subjects) mm3 resolution. Cortical parcellation
of the high-resolution T1 image was done in Freesurfer (Massachusetts General Hospital, Boston,
Boston). Human motor association area template (HMAT) was implemented in order to define the
ROIs that are specifically related to human motor function [Mayka et al., 2006]. For the test of
reproducibility, thirty-six cortical ROIs were selected as the purpose was to quickly examine
network measures initially. For studies in PD and MS, sixty-eight cortical and subcortical ROIs
were included for connectivity analysis. We considered as many cortical regions as possible but
excluded the ROIs that may have a poor signal-to-noise ratio (SNR) in fMRI due to partial volume
effect such as the frontal pole, temporal pole, etc. Table 4.3 lists these 68 ROIs. A brain mask was
applied before registration to remove non-brain tissue and we registered the structural image to the
mean fMRI image. The same transformation was subsequently applied to the ROI mask in order
to obtain the ROIs in fMRI resolution. The ROIs acted as masks to determine the appropriate
voxels making up the average ROI time courses. Finally, the fMRI time courses of selected ROIs
were extracted using in-house scripts in Matlab and the data were detrended before connectivity
analysis. Subjects who had translational and rotational head movements during data acquisition of
more than 2 mm and 2 degrees respectively were excluded.
4.2.4 Functional connectivity analysis
In the 11 healthy subjects, binary Pearson’s r correlation matrices were created to form the
undirected/unweighted “edges” and the 36 ROIs acted as the “nodes”. The Brain Connectivity
Toolbox (https://sites.google.com/site/bctnet/) [Rubinov and Sporns, 2010] was used to calculate
graphical measures including global efficiency, assortativity, characteristic path length, density,
97
information flow coefficient, modularity, and rich-club coefficient, which were commonly used in
clinical research. Table 4.4 describes the definition of the graphical measures used in this chapter.
Repeated measure ANOVA (rmANOVA) and was applied to test whether these measures were
significantly different across three scans and coefficient of variation (COV) (standard deviation
divided by the mean) was calculated to evaluate the variance across time.
Bilateral ROIs
thalamus middle temporal gyrus
pallidum superior temporal gyrus
caudate superior occipital gyrus
hippocampus fusiform gyrus
amygdala lingual gyrus
insula inferior parietal gyrus
accumbens postcentral gyrus
superior frontal gyrus posterior cingulate cortex
rostral middle frontal gyrus precuneus
caudal middle frontal gyrus superior parietal gyrus
inferior frontal gyrus angular gyrus
lateral orbitofrontal cortex supramarginal gyrus
medial orbitofrontal cortex cerebellum cortex
caudal anterior cingulate cortex primary motor cortex
rostral anterior cingulate cortex supplementary motor area
entorhinal pre supplementary motor area
inferior temporal gyrus
Table 4.3 68 ROIs are used in the PD and MS studies.
Measures Definition
global
measures
global efficiency the average inverse shortest path length in the network (measure of
integration)
transitivity the ratio of triangles to triplets in the network (measure of
segregation)
modularity it quantifies the degree to which the network may be subdivided into
clearly different groups (measure of segregation)
98
assortativity
correlation coefficient between the degrees of all nodes on two
opposite ends of a link (nodes link to other similar nodes, measure of
resilience) characteristic path
length
the average shortest path length in the network (measure of
integration)
rich club
coefficient
the fraction of edges that connect nodes of degree k or higher out of
the maximum number of edges that such nodes might share (in this
study k = 6 in PD & 10 in MS)
local
measures
betweenness
centrality
the fraction of all shortest paths in the network that contain a given
node (measure of centrality/hub)
local efficiency the global efficiency computed on the neighborhood of the node
(measure of segregation)
Table 4.4 Definitions of the graphical measures used in this study.
In PD and MS studies, partial correlation analysis was conducted to generate a connectivity matrix
for each subject, which resulted in a 68-by-68 matrix. The Brain Connectivity Toolbox (BCT)
[Rubinov and Sporns, 2010] was used to compute graph theoretical measures. The partial
correlation matrix was proportionally thresholded and binarized with the density of 15% to ensure
equal density across subjects [van den Heuvel et al., 2017]. For global measures, global efficiency,
transitivity, modularity, assortativity, characteristic path length, rich club coefficient (for PD, it
was at level 6, which is the highest degree that did not give a Not-a-Number (NaN) across subjects
due to matrix sparsity; for MS, it was at level 10) were computed [Fornito et al., 2016; Rubinov
and Sporns, 2010]. These global measures summarized network characteristics of the entire brain
network. For local measures, betweenness centrality and local efficiency were computed [Fornito
et al., 2016; Rubinov and Sporns, 2010], which reported nodal characteristics in the network. Two
99
sample t-tests were carried out to test whether graph theoretical measures were significantly
different between patient and control groups.
4.2.5 Brain-behaviour analysis
Several approaches were applied to study the brain-behaviour relationship (i.e. associations
between graph theory measures and cognitive scores) in PD and MS studies only.
First, we utilized Spearman’s correlation analysis exploring whether individual cognitive tests
were correlated with graph theoretical measures in PD and MS. The correlation analyses were
carried out on all the global measures against raw cognitive scores from individual tests. For local
measures, only ROIs that showed significant differences in t-tests were included in the analyses.
The local measures of these ROIs were included to correlate with raw cognitive scores. Bonferroni
correction was applied to correct for multiple comparisons. Of note, only composite scores were
included in the brain-behaviour analysis as these scores were representative for cognitive domains.
For example, only WMI and PSI were included rather than all subscores of WAIS IV.
Linear regression was carried out on all global measures against cognitive scores. In other words,
all global measures were concatenated in a matrix and acted as predictors, while each cognitive
score was the response variable. The p values of the linear regression model and R-squared values
between real and predicted cognitive scores were reported. Finally, we performed LASSO
regression on local measures (predictor data) with 10-fold cross validation to select the nodal
values which contributed to individual behavioural scores (the responses). The regression
coefficients for such LASSO model were then included in a linear regression to “predict”
behavioural scores. At this stage, all calculations were done with all the subjects as a whole without
100
separating data into training and testing data sets; therefore, we referred the findings as “regression
results”. Results with R-squared higher than 0.5 were considered good results and p values were
reported. This preliminary approach allowed selection of variables between graphical measures
and cognitive scores. Finally, we repeated such processes with training and testing data sets to
precisely predict the behavioural outcomes in a leave-one-out fashion, which were referred as
“prediction results”.
4.3 Results
4.3.1 Reproducibility
None of the graphical measures showed significant differences among the three fMRI sessions (pr
> 0.05 in all measures, Figure 4.1). Moreover, COV of most of the measures derived from three
fMRI sessions demonstrated similar values except the measures that were non-Gaussian
distributed (Table 4.5). The results indicate that these measures are reproducible across scans.
Figure 4.1 rmANOVA results of graphical measures across 3 fMRI sessions
101
Graphical Measures 1st fMRI 2nd fMRI 3rd fMRI
global efficiency 42.7% 29.4% 43.2%
assortativity 70.3% # 70.3% # 61.6% #
characteristic path length 33.4% 29.4% 32.3%
density 17.8% 16.2% 16.5%
information flow coefficient 31.5% 28.4% 30.2%
modularity 6.8% 3.9% 9.6%
rich club coefficient 36.4% * 31.1% * 18.8%
Table 4.5 COV of all graphical measures across fMRI sessions.
* not Gaussian
# take absolute values
4.3.2 Parkinson’s Disease
4.3.2.1 Graphical measures
None of the global graph theoretical measures were different between PD and HS, however,
several ROIs showed altered local efficiency and betweenness centrality. Figure 4.2 left panel
shows the average connectivity patterns in PD on a brain template. Color-coded connections
indicated the connectivity within the ROIs that distinguish PD and HS. Figure 4.2 right panel
presents the local measures in PD on a ring diagram. The right hippocampus, left supramarginal
gyrus, left pre-motor cortex, left middle temporal gyrus, right entorhinal cortex, left postcentral
gyrus, left amygdala, left angular gyrus, and right postcentral gyrus demonstrated higher local
efficiency in PD (p<0.05, uncorrected) (Figure 4.2, Appendices C.1), indicating that functional
connectivity was more segregated in PD. Figure 4.3 left panel highlights the connections that
differentiate PD and HS. Figure 4.3 right panel demonstrates local measures in HS ranked by
betweenness centrality in a ring diagram and color-coded the ROIs that were significantly different
102
between two groups. These ROIs showed lower betweenness centrality in PD such as the left
superior frontal gyrus, bilateral superior parietal cortices, left middle temporal gyrus, and right
inferior frontal gyrus; while the right accumbens area and right pallidum exhibited higher
betweenness centrality in PD (Figure 4.3, Appendices C.1). The changes of betweenness centrality
in PD indicated an altered hub organization, where important nodes lost significance and the nodes
with a less central role have become more important. However, the results did not survive for
multiple comparison. Appendices C.2 shows another analysis with logistic LASSO to indicate that
graphical measures can be clearly separated into PD and HS groups.
103
Figure 4.2 Local graphical measures in PD. (left panel) Partial correlation connectivity patterns with the
6.32th percentiles (strongest 144 connections) are shown in PD. The orange connections indicate the
connectivity within the ROIs that present higher local efficiency in PD. (right panel) Local efficiency (red
bars) and betweenness centrality (black bars) of each node ranked by local efficiency are shown in a ring
diagram. Bold font size indicates the regions that show higher local efficiency in PD (p<0.05). The figure is
derived from NeuroMArVL (http://immersive.erc.monash.edu.au/neuromarvl/).
104
Figure 4.3 Local graphical measures in HS. (left panel) Partial correlation connectivity patterns with the
6.32th percentiles (strongest 144 connections) are shown in HS. The orange connections indicate the
connectivity within the ROIs that distinguish PD and HS with higher betweenness centrality in PD; while
the blue ones represent the connections between the ROIs that differentiate two groups with decreased
betweenness centrality in PD. (right panel) The ring diagram presents betweenness centrality (black bars)
and local efficiency (red bars) of each node ranked by betweenness centrality. Bold and bigger font size
indicates the regions that are significantly different between two groups (p<0.05). Except the right pallidum
and accumbens, all the regions with bold font show lower betweenness centrality in PD. The figure is derived
from NeuroMArVL (http://immersive.erc.monash.edu.au/neuromarvl/)
105
4.3.2.2 Brain-behaviour association
In the Spearman’s rank correlation analysis, only raw Symbol Digit Modality Test (SDMT) scores
were significantly correlated with global efficiency and characteristic path length (p = 0.002 and
rs = -0.6, p = 0.002 and rs = 0.61, respectively, survive for Bonferroni correction with 14 tests were
run) (Figure 4.4). Higher global efficiency was related to lower SDMT scores; while higher
characteristic path length (which is the inverse measure of global efficiency) was associated with
better performance. Overall, correlation results indicated that only specific network measures were
correlated with the performance of certain cognitive tests. No significant associations were found
in local measures.
Figure 4.4 Spearman’s correlation between cognitive scores and graphical measures in PD. Spearman's
correlation shows that SDMT is negatively correlated with global efficiency and positively correlated with
characteristic path length. These correlations survive for Bonferroni correction.
[SDMTTOT: Symbol Digit Modalities Test total scores, chapath: characteristic path length]
106
With all global measures as predictors and individual behavioural scores as responses, linear
regression did not reveal any significant models for PD. While performing LASSO on local
measures to predict behavioural outcomes, the models were over-fitted as there were 23
observations but 68 features in one model even the processes were carried out in leave-one-out
fashion. Therefore, we were unable to report results for regression and prediction analysis in PD.
However, with a multivariate approach, better cognitive performance was related to higher
modularity and transitivity. The analysis and details are shown in Appendices C.4.
4.3.3 Multiple Sclerosis
4.3.3.1 Graphical measures
None of the global and local measures showed significant differences between MS and NC.
4.3.3.2 Brain-behaviour association
In the Spearman’s rank correlation analysis, only the performance of the Verbal Fluency Test
(FAS) was significantly correlated with modularity in MS (r= 0.49, p=0.001, survived for
Bonferroni correction with 14 tests as indicated in section 4.2.1) (Figure 4.5). Higher FAS score
(better performance) was related to higher modularity (i.e. more segregated networks).
107
Figure 4.5 Spearman’s correlation between cognitive scores and graphical measures in MS. Spearman's
correlation shows that FAS is positively correlated with modularity. Such correlation survives for
Bonferroni correction.
As none of the local measures demonstrate differences, we did not included either betweenness
centrality or local efficiency in the Spearman’s rank correlation analysis. In linear regression, the
model of all global measures against FAS appeared significant (p=0.04); however, only modularity
demonstrated significant weighting (p=0.0038) and the predictability of this model was small (R-
squared=0.16). In other words, FAS was correlated with modularity, but modularity cannot
accurately predict FAS in a linear regression.
In the regression results of local measures, we discovered 6 models that demonstrated significant
relations between LASSO-selected nodal measures and behavioural outcomes. Selected
betweenness centrality exhibited significant relations with EDSS (p<0.001, R-squared=0.66),
108
MDITOT (p<0.001, R-squared=0.65), and transformed TMTB (p<0.001, R-squared=0.82) (Figure
4.6, Table 4.5). Selected local efficiency showed significant relations with DD (p<0.001, R-
squared=0.77), PSI (p<0.001, R-squared=0.89), and STAIS (p<0.001, R-squared=0.77) (Figure
4.6, Table 4.5). Furthermore, in the prediction results, whereby 23 training and 23 testing data sets
were applied, 8 nodal measures of local efficiency accurately predicted EDSS with p=0.0078 and
R-squared = 0.55 (Figure 4.7). These 8 nodal measures were the local efficiency of the following
ROIs: left thalamus, left inferior frontal gyrus, left lateral orbitofrontal gyrus, left entorhinal cortex,
right accumbens area, right lateral orbitofrontal gyrus, right primary motor cortex, and right
supplementary motor area.
Local
measures
LASSO-selected ROI (gyrus/cortex) Behavioural
score
Statistics
betweenness
centrality
L caudal middle frontal, L rostral anterior cingulate, L
inferior temporal, L middle temporal, L inferior parietal, L
M1, R caudate, R lateral orbitofrontal, R middle temporal, R
superior occipital, R superior parietal
EDSS
R2 = 0.66
p=3.0e-
07
betweenness
centrality
L inferior frontal, L lateral orbitofrontal, L superior temporal,
L Lingual, L Angular, M1, R thalamus, R rostal middle
frontal, R medial orbitofrontal, R inferior temporal, R
superior occipital, R cuneus, R superior parietal, R
supramarginal, R preSMA
MDITOT
R2 = 0.65
p=7.7e-
06
109
betweenness
centrality
L Hippocampus, L rostral middle frontal, L lateral
orbitofrontal, L middle temporal, L superior temporal, L
cerebellum, L preSMA, R hippocampus, R superior frontal,
R rostral middle frontal, R inferior frontal, R lateral
orbitofrontal, R inferior temporal, R supramarginal
transformed
TMTB
R2 = 0.82
p=7.3e-
10
local
efficiency
L thalamus, L hippocampus, L inferior frontal, L entorhinal,
L cerebellum, L M1, R hippocampus, R accumbens, R
superior frontal, R caudal middle frontal, R lateral
orbitofrontal, R medial orbitofrontal, R rostral anterior
cingulate, R entorhinal, R superior temporal, R lingual, R
M1, R SMA
disease
duration
R2 = 0.77
p=2.1e-
07
local
efficiency
L hippocampus, L inferior frontal, L medial orbitofrontal, L
caudal anterior cingulate, L rostral anterior cingulate, L
inferior temporal, L fusiform, L inferior parietal, L
postcentral, L post cingulate, L angular, L supramarginal, L
M1, L SMA, L preSMA, R caudate, R insula, R accumbens,
R superior frontal, R lateral orbitofrontal, R rostral anterior
cingulate, R fusiform, R lingual, R inferior parietal, R
postcentral, R cuneus, R cerebellum
PSI
R2 = 0.89
p=1.6e-
07
local
efficiency
L hippocampus, L caudal middle frontal, L entorhinal, L
superior temporal, L superior occipital, L fusiform, L
angular, R thalamus, R pallidum, R caudate, R hippocampus,
R amygdala, R superior frontal, R rostral middle frontal, R
caudal anterior cingulate, R entorhinal, R middle temporal,
R superior temporal, R inferior parietal, R postcentral, R
preSMA
STAIS
R2 = 0.77
p=3.3e-
06
110
Table 4.6 Regression results of local measures in MS. The table shows the statistical power to predict
behavioural scores using the nodal measures selected by Lasso regression.
[L: left hemisphere, R: right hemisphere, M1: primary motor cortex, preSMA: pre supplementary motor area,
SMA: supplementary motor area, EDSS: Expanded Disability Status Scale, MDITOT: Multiscore Depression
Inventory Total Scores, TMTB: Trail Making Test B, PSI: Processing Speed Index, STAIS: State-Trait Anxiety
Inventory State]
Figure 4.6 Regression results of local measures in MS. The LASSO-selected nodal measures are included in
a linear regression to calculated scores and the predictability (p and R2 values) is reported.
111
Figure 4.7 Prediction results of local
measures in MS. The 8 LASSO-selected
nodal measures of 23 training data sets
are included in a linear regression to
predict behavioural scores in 23 testing
data sets.
4.4 Discussion
Our results of reproducibility further support previous studies that have shown that graphical
measures are robust observations for resting state functional connectivity. Although sophisticated
statistical methods shall be implemented such Inter Class Correlation (ICC), our preliminary
results reinforced the rationale of using graphical measures to study brain organization.
4.4.1 Changes of brain organization
Overall brain organization can be thought of as a balance between integration and segregation, in
which the former facilitates information integration across the whole brain; while the latter enables
information transfer within individual networks [Sporns, 2013]. This organization allows the brain
to function in an economical way by reducing the wiring cost of linking anatomically segregated
regions [Bullmore and Sporns, 2012]. In addition, hub regions – the cortical areas that play a
112
central role in the networks – support information integration in many cognitive functions as well
as neuronal coupling between networks [van den Heuvel and Sporns, 2013]. These principles
together maintain brain function and support cognition in healthy subjects; in the diseased brain,
this delicate balance is altered and fails to sustain normal functioning [Crossley et al., 2014; Fornito
et al., 2015; Stam, 2014].
Some studies have reported altered functional organization measured by graph theory in PD. For
example, PD with MCI presented increased measures of segregation such as modularity and
clustering coefficient and these measures in the frontal areas were both negatively and positively
correlated with cognition, meaning that functional connectivity in the frontal regions has become
more segregated and failed to support cognitive functions [Baggio et al., 2014]. Although other
studies reported decreased clustering coefficient and local efficiency [Luo et al., 2015a], which
implied that FC might be less segregated, differences in methodology should be taken into account.
Increased local efficiency has been observed in PD in another study, further suggesting that
information transfer in PD is better within local sub-networks than at the global network level
[Berman et al., 2016]. Furthermore, hub reorganization has been described, in which PD lost hubs
but other non-hub regions became more important as compensation [Koshimori et al., 2016].
Our results of local graph theoretical measures indicated increased segregation of rsFC in PD,
consistent with the previous studies utilizing graph theory. As ROIs across temporal, occipital,
parietal, motor association and sensory regions presented higher local efficiency in PD, we propose
that functional connectivity of multiple regions in PD has become more segregated than that in
HS. Interestingly, most of the ROIs that showed significant differences also had higher local
efficiency in PD (Figure 4.2), implying that these regions played important roles in local networks
113
rather than global connectivity. We also observed hub changes in this cohort, which demonstrated
decreased betweenness centrality (i.e. an index of hub and measures how important a given node
is) in the ROIs that showed higher values than other regions. These ROIs have been previously
defined as functional hubs, such as the superior frontal gyrus, superior parietal gyrus, middle
temporal gyrus, and inferior frontal gyrus [van den Heuvel and Sporns, 2013] (Figure 4.3).
Although the results were not apparent after controlling for multiple comparison, there was still a
trend that these hub regions were affected compared to non-hub regions. In addition, as shown in
Appendices C.3, with a more nuanced approach (i.e. Logistic LASSO), local measures can be
clearly categorized into PD and HS groups, implying that collectively these nodal measures
provide a robust way to distinguish between groups. Thus our findings are consistent with the
vulnerability of hubs in neurological disease populations [Crossley et al., 2014; Stam, 2014]. Only
two ROIs (the right accumbens and pallidum) in this PD population demonstrated increased
betweenness centrality and these ROIs in HS showed the smallest values (Figure 4.3).
The increase in connectivity in the right accumbens and pallidum that we observed is intriguing.
The nucleus accumbens (NAc) receives dopaminergic projections from the ventral tegmental area
(VTA) and substantia nigra (SN) regions and projects to several deep grey matter areas which
include globus pallidus; therefore, the NAc has been hypothesized to be associated with the
nigrostriatal and mesolimbic systems [Salgado and Kaplitt, 2015] and linked to reward behaviour
[Knutson et al., 2007]. In PD, the NAc receives less dopamine projection from the VTA and SN
and the globus pallidus receives output from the NAc, so the increased FC is perhaps surprising.
The PD patients in this study were relatively mildly affected, and likely capable of a significant
114
amount of compensation. Thus NAc and pallidum may up-regulate their overall connectivity to
maintain function.
Unlike previous studies which reported changes in global measures such as modularity [Baggio et
al., 2014], we did not observe differences in global indices. Again, as our cohort was relatively
mildly affected, the limited damages in local regions might not be severe enough to impact global
measures. Local FC might be more sensitive to pathological and physiological features and
potentially serve as biomarkers than global measures at early disease stages.
In our MS cohort, we did not find any graphical measures which were significantly altered.
Although the sample size was relatively larger than that in our PD cohort, none of the MS subjects
demonstrated cognitive impairments at the time of examination. Therefore, we suspected that the
pathological damages were not severe enough to impact functional connectivity measured by
graph theory. However, brain-behaviour analysis revealed some promising results (discussed in
the following sections), which implied that even though there were no significant changes of brain
organization, subtle alterations of graphical measures were informative of cognitive decline and
disease severity.
4.4.2 Reduced functional integration and increased functional segregation correlated
with better cognitive performance
The relations between FC, graph theoretical measures, and cognition have gained attention lately
in network neuroscience and clinical studies. Several graphical measures have been associated
with intelligence, working memory, and executive functions in healthy subjects [Cohen and
D’Esposito, 2016; Pamplona et al., 2015; Reineberg and Banich, 2016], yet research in
115
neurological disorders has been limited to a few studies. With a correlation approach, a study
reported that hub organization in PD was more related to dopaminergic medication dosage rather
than cognitive functions [Koshimori et al., 2016]. Although the study did not rule out the
associations between hubs and cognitive functions that a simple correlation method might not be
able to catch, more sensitive statistical analyses are required to explore complex relations between
cognition and FC. With a linear regression analysis, Baggio et al. discovered that increased local
measures in PD with MCI, such as clustering coefficient, local efficiency, and modularity in the
frontal areas, were correlated with worse attention and executive function [Baggio et al., 2014].
As previously mentioned in the introduction (section 4.1.2), some studies have proposed that
function connectivity in MS is more segregated (i.e. reduced integration) compared to healthy
subjects [Gamboa et al., 2014; Liu et al., 2017; Rocca et al., 2016a]; however, whether such
segregated brain organization is related to any cognitive function remains unclear. Overall, in both
PD and MS, more evidence of the associations between cognitive functions and FC is needed to
underline the neural mechanism of cognition and how the disease affects behaviour.
In this study, we only observed strong correlations between measures of integration (i.e. global
efficiency and characteristic path length) and SDMT scores in PD with a correlation analysis. The
performance of SDMT requires attention, processing speed, scanning abilities, and engages several
brain regions including the occipital cortex, middle frontal gyrus, precuneus, superior parietal
lobes, and cerebellum [Forn et al., 2011] and has been proven as a robust tool to detect cognitive
impairments in healthy aging and neurological disorders [Alamri et al., 2017; Parmenter et al.,
2007; Sheridan et al., 2006]. Therefore, it would be reasonable to assume that higher functional
integration is related to better SDMT performance. In fact, our results demonstrated the opposite
116
trend, whereby higher functional integration (i.e. higher global efficiency and lower characteristic
path length, as they are inversely related) was related to poor performance of SDMT. Furthermore,
as shown in Appendices C.3, with a multivariate approach we revealed that better cognitive
function in several domains was associated with higher measures of functional segregation,
supporting the results with the univariate method (i.e. correlation). Such paradoxical relations may
be potentially due to disease effects, leading toward a more segregation-oriented brain organization
in order to respond to cognitive demand. This is consistent with prior reports, demonstrating more
segregated FC in frontal areas related to attention/executive function in PD [Baggio et al., 2014].
We suggest that “pathological resonance” in the basal ganglia–cortical network, previously
described in PD [Eusebio et al., 2009], may result in excessive regional integration but overall
global segregation.
Although increased modularity in MS has been related to worse dual task performance [Gamboa
et al., 2014], we, in fact, observed the opposite relation. Higher modularity, which implies a
stronger subdivision into segregated groups of nodes, was correlated with better performance of
verbal fluency test in our cohort. Such association asserted that the executive skills of
spontaneously generating information according to rules are associated with brain regions
subdividing into segregated groups in mild MS. Patient demographics could be a factor to explain
these discrepancies between studies. First, the previous study [Gamboa et al., 2014] included both
RRMS and clinically isolated syndrome (CIS) patients as they counted for “early stage of MS”.
On the other hand, we only included RRMS in the study. Moreover, the sample size in our study
is almost 6 times bigger (46 RRMS in our study, 8 RRMS in the previous study) than that in the
previous research, which provides stronger statistical power to obtain robust results. Finally,
117
although the dual task included PASAT and a maze test, subjects were instructed to give priority
to the PASAT. The cognitive performance of the previous study required different skills compared
to the Verbal Fluency Test. Presumably, such dual task requires more attention, calculation,
processing speed, and visuospatial orientation abilities; while the Verbal Fluency Test measures
higher-order skills which is more goal-oriented. Therefore, we concluded that better executive
functioning is associated with more segregated rsFC in MS. Such segregated rsFC, which
represents a reduction of network efficiency, could be a result of compensating for tissue damage
before network collapse [Fleischer et al., 2017].
The optimal balance between brain segregation and integration for higher-order cognitive function
remains a source of debate, with previous research arguing for more of one or the other [Cohen
and D’Esposito, 2016; Reineberg and Banich, 2016]. Between-network communication (i.e.
integration) may be important for working memory [Cohen and D’Esposito, 2016], but executive
functions may require more nodal FC (i.e. segregation) [Reineberg and Banich, 2016]. The delicate
balance of segregation/integration may relate to the balance between focusing on internal brain
states versus external sensory input [Miller and Cohen, 2001]. Sensory input from several cortices
can send signals to the prefrontal cortex, which integrates this information with internal brain
states, with the resultant output transferred to subcortical and motor association areas to executive
the action. Presumably, during processing of input and output in the prefrontal regions, integration
is required in order to communicate with different brain regions; while processing internal states,
redundant and unnecessary regions are excluded in order to focus on the processes within the
prefrontal cortex. We propose that the disease may have preferentially affected integration between
remote regions rather than internal processing, consistent with the “pathological resonance”
118
concept alluded to above in PD. Moreover, in MS, segregated brain organization may represent a
compensatory mechanism before network collapse.
4.4.3 Predictability of graphical measures to behaviour
One of the challenges to predict behavioural outcome using imaging data is sample size.
Unfortunately, due to small sample size, we were unable to perform LASSO analysis on local
measures in PD as there were too many features and not enough observations. The linear regression
model did not show significant relation between global measures and behavioural outcome in PD.
Therefore, this section mainly discusses the findings in MS.
Although correlation analysis showed the associations between individual graphical measures and
behavioural scores, further analyses can be done to investigate whether any of the features were
jointly related to any behavioural outcome. In this study, with linear regression, only modularity
was related to verbal fluency performance, further supporting the relation between segregated
brain organization and executive skills. However, the predictability of such altered rsFC to
executive skills was not promising in linear regression as R-squared value between real and
predicted FAS was only 0.16. Taken together, certain executive skills were indeed related to
segregated brain networks (i.e. brain networks have more modules rather than working as a whole),
but the association was not strong enough for brain connectivity to accurately predict behaviour.
For local measures, as there were too many features for linear regression and caused insufficient
observations, we decided to perform LASSO first as a step of feature selection. The selected nodal
graphical measures were included as predictor data in linear regression and individual scores acted
as responses. This process revealed which nodal measures were influential to behavioural
119
outcomes and whether such influence was strong enough to predict individual scores. For EDSS,
the betweenness centrality (indicate the importance of a given node) across frontal, temporal,
parietal, and occipital regions as well as motor areas demonstrated strong impacts to predict disease
severity measured by physical disability. For depression symptoms, interestingly, most of the
influential nodes were located in the frontal regions and a subset were in the occipital, parietal,
and subcortical areas. For TMTB performance, the importance of several frontal and temporal
nodes were significantly influential as well as the inferior parietal region and cerebellum. These
nodal measures together highly predicted TMTB performance, which requires task switching
ability. Therefore, we propose that 1) the disease severity was related to distributed regions in MS,
2) the importance of frontal regions was largely related to depression symptoms in MS, and 3)
executive function (i.e. task switching ability) was significantly associated with the coordination
between not only frontal, temporal, and parietal regions but also the cerebellum, supporting the
“cognitive role” of cerebellum in neurological diseases [Buckner, 2013].
Local efficiency (i.e. network efficiency in nodal level) of frontal, subcortical, and motor-
associated regions highly predicted disease duration. Of note, this was the only model that could
predict disease duration in the testing data. In the prediction results, even though fewer nodes were
selected, these regions still covered frontal, subcortical, and motor association regions, which
further emphasized the role of these regions regarding disease duration. The longer the disease
duration, the higher nodal efficiency in these regions. For processing speed ability, highly
distributed regions were included, highlighting that processing speed requires distributed regions
across whole brain (include cerebellum) to coordinate together. Finally, unlike depression, state
anxiety was more related to nodal efficiency in subcortical and temporal regions rather than the
120
frontal cortex, suggesting a different mechanism compared to depression even though both
symptoms are affective. To conclude, pathology-related scores were more associated with nodal
efficiency in frontal, subcortical, and motor areas; while cognition-related scores were more
associated with connectivity across whole brain. Anxiety, on the other hand, involved different
mechanism opposite to other affective disorders.
4.4.4 Limitations
First, due to motion and image artifacts, eight PD subjects had to be removed from the connectivity
analysis and the sample size was relatively small. For advanced statistical approaches, including
machine learning methods, a bigger sample size will likely be needed. Although the sample size
of MS cohort seemed to be sufficient to preform LASSO regression and leave-one-out cross
validation was applied, a bigger sample size would still be required to maximize the advantages of
machine learning approaches and avoid over-fitting. Moreover, subjects in both study cohorts
tended to be mild and were in the early stages of the disease. This could be the reason why we did
not observe significant differences between patients and HS after correcting for multiple
comparison and why we found few associations between graphical measures and cognitive scores
in correlation and regression analyses.
4.5 Conclusion
In this study, with a graph theoretical approach, we demonstrated that rsFC was more segregated
in PD across regions and PD subjects demonstrated hub vulnerability. Increased connectivity in
the nucleus accumbens and pallidum suggested possible compensation for PD pathologies in
mildly-affected individuals. With a correlation analysis, we concluded that attention and
121
processing speed abilities were associated with segregated FC in PD, possibly related to
pathological synchrony in basal ganglia structures. This conclusion was further supported by the
results of multivariate analyses. In MS, the ability of generating information based on rules was
also correlated with segregation rsFC, possibility representing a compensatory effect before
network collapse. Furthermore, nodal graphical measures across frontal, subcortical, and motor
association areas can predict disease progression; while nodal measures across cortices in
cerebrum and cerebellum predicted executive skills. Finally, different affective disorders in MS
may involve different mechanisms.
122
Chapter 5: Dynamic functional connectivity and executive function
This chapter investigates whether time-varying dynamic functional connectivity (dFC) at rest can
be used as a biomarker for neurological disorders (i.e. PD and MS). Several dynamic features are
calculated to summarize how connectivity changed over time. Moreover, how dynamic features
of rsFC are related to cognition, especially executive function, and reproducibility of these
dynamic features is also investigated.
5.1 Introduction
5.1.1 Dynamic functional connectivity
Recent evidence has shown that connectivity fluctuates across time from seconds to minutes even
in the resting state, which can be estimated by models of dynamic functional connectivity (dFC)
[Allen et al., 2014; Betzel et al., 2016; Chang and Glover, 2010; Handwerker et al., 2012;
Hutchison et al., 2013a; Jones et al., 2012]. The simplest, and perhaps most common time-varying
approach to assess dFC is to estimate correlations between brain regions within a fixed-length,
sliding window, with the (possibly overlapping) windows ultimately moved over the entire data.
Nevertheless, there are potential pitfalls with such an approach [Hindriks et al., 2016; Hutchison
et al., 2013b]; if the window is too long, then important dynamic changes may be missed. If the
window is too short, the connectivity estimates may be unstable as too few samples are available
for the statistical inferences. A window length of 30-60 seconds for fMRI data has been
heuristically suggested [Leonardi and Van De Ville, 2015; Zalesky and Breakspear, 2015]
123
5.1.2 Dynamic functional connectivity and cognition
Dynamic functional connectivity (dFC), also referred to as network dynamics and assessed by
time-varying approaches, appears to be particularly pertinent to several cognitive processes
including memory, language, attention, and executive functions [Braun et al., 2015; Bressler and
Scott Kelso, 2001; Kucyi et al., 2016; Mattar et al., 2015; McIntosh et al., 2008; Nomi et al., 2017;
Shafto and Tyler, 2014; Thompson et al., 2013]. Increased dynamical variability in the EEG was
found to be correlated with better performance (i.e. shorter reaction time and higher accuracy) in
a memory task, emphasizing the importance of brain complexity in cognitive development
[McIntosh et al., 2008]. There are associations between dynamic changes in
frontoparietal/frontotemporal networks and neuropsychological measures, showing that the
flexibility of neuronal activity in the frontal regions is cognitively beneficial for working memory
performance and executive functioning [Braun et al., 2015]. This has led to a proposed “functional
cartography” of the cognitive system, based on the estimated amount of integration and
recruitment of brain regions during different cognitive processes [Mattar et al., 2015]. In addition,
cognitive flexibility, which is an important part of executive function, has been associated with
greater variability of resting-state connectivity [Nomi et al., 2017]. Network dynamics may also
be important for language function in an aging population [Shafto and Tyler, 2014]. Measures of
dFC have been recently applied to understand how the human brain is affected by diseases such as
Parkinson’s disease, Schizophrenia, Alzheimer's disease, and major depression [Damaraju et al.,
2014; Kaiser et al., 2015; Madhyastha et al., 2014; Sakoglu et al., 2010; Wee et al., 2013].
124
5.1.3 Dynamic functional connectivity in diseased populations
A sliding window approach combined with k-means clustering, demonstrated that patients with
schizophrenia have shorter “dwell time” in metastable states [Damaraju et al., 2014], implying an
unstable connectivity pattern. In schizophrenia, decreased connectivity between subcortical
regions and sensory networks were only observed in dynamic networks, suggesting that static
connectivity was less sensitive to functional abnormalities. Research in major depression has
revealed decreased dynamic resting-state functional connectivity (drsFC) between the medial
prefrontal cortex and parahippocampus, and increased dynamic connectivity between the medial
prefrontal cortex and dorsolateral prefrontal cortex [Kaiser et al., 2015]. These distinct patterns
could be the results of positive and negative correlations in activity across sliding windows, which
would not have been captured in static functional connectivity analysis alone. Alzheimer’s disease
patients with mild cognitive impairment (MCI) exhibit altered graphical measures such as
decreased small-world coefficients and smaller clustering coefficients in some temporal networks
[Wee et al., 2013], indicating a failure to maintain a small-world brain connectivity compared to
healthy subjects.
In Parkinson’s disease, with a sliding window approach, altered dFC in the dorsal attention and
frontoparietal network was related to performance in an attention task [Madhyastha et al., 2014].
With a sliding window approach combined with Hilbert transform, deep-brain stimulation (DBS)
has been shown to “rebalance” the global dynamics in PD towards a healthy regime in the
thalamus, globus pallidus, and right orbitofrontal regions judged by “phase consistency” [Saenger
et al., 2017]. Although this study did not investigate the relations between DBS-altered dFC and
cognitive function, the findings showed that dFC can be an index of disease progression.
125
Combining graph theory, independent component analysis, the sliding window approach, and
clustering methods together, PD subjects have shown altered rsFC states, whereby networks were
more strongly interconnected between each other rather than sparsely connected, indicating that
connections were stronger but less dynamic in PD [Kim et al., 2017]. Such connectivity pattern
caused abnormal functional integration with higher variability, indicating an inefficiency of
information transfer in PD [Kim et al., 2017]. A similar approach was applied to PD patients with
MCI and without MCI. Only patients with MCI demonstrated altered dFC, whereby PD-MCI spent
less time in the “hypo-connectivity” state and showed more state transitions, implying that PD-
MCI actually presented weaker but more dynamic rsFC than PD and normal subjects [Díez-Cirarda
et al., 2018]. Perhaps, taken together, stronger rsFC becomes more rigid but weaker rsFC becomes
more dynamic in PD. On the other hand, with a similar approach, patients with MS demonstrated
the opposite trend compared to PD.
MS patients with MCI spent less time in a “high-connectivity” state and demonstrated less switches
between state transitions (i.e. lower dynamic fluidity) [D’Ambrosio et al., 2018]. Another study
applied principal component analysis (PCA) to identify “eigenconnectivity” after data were
processed in the sliding window approach [Leonardi et al., 2013]. People with MS showed reduced
strength of drsFC in distributed regions including the amygdala, occipital, parietal, middle and
posterior cingulate, and superior frontal regions. Moreover, two of the eigenconnectivity
components demonstrated differences between MS and healthy subjects, whereby MS had stronger
contributions than controls (i.e. the connectivity patterns were more obvious in MS). One
component showed stronger mean drsFC in the posterior part of the default mode network (DMN),
while the weaker mean drsFC in anterior and middle temporal regions. The other component
126
demonstrated stronger mean drsFC in temporal and angular areas; while weaker mean drsFC was
found in fronto-parietal, right amygdala, and motor regions [Leonardi et al., 2013]. Although the
study revealed that dFC can be a functional biomarker for MS, no correlations were reported
between connectivity patterns and clinical data. To conclude, these studies have concluded that
dFC has functional relevance and provides further insights into how brain networks are affected
in neurological disorders. However, robust associations between dFC and cognitive domains,
especially executive functioning, rather than cognitive states (i.e. cognitive impaired vs cognitive
intact) remain absent.
In this study, we utilize a sliding window approach to calculate connectivity differences across
time and quantify these changes by estimating dynamic features. Furthermore, methods of
correlation and regression are implemented to explore the relations between dFC and cognitive
function in PD and MS. In addition, similar to the last chapter, we carry out an analysis to test
whether the dynamic features are reproducible across time.
5.2 Materials and methods
5.2.1 Subjects and behavioural data
Eleven healthy subjects (mean age: 25.7±3.6, 4 females, 7 males, all with university education
level) were recruited for the test of reproducibility as section 4.2.1.
PD data are from research project GFM2 and MS data are from COGMS project. The MS cohort
is the same as chapter 2 and 4, but PD cohort is different from the previous chapters. A subset of
MS subjects are used for comparison with age-matched healthy subjects as previous chapter.
127
The same groups of MS subjects and normal control subjects (NC) as the last chapter were included
(46 MS for correlation and regression, 18 MS and 15 NC for comparison; details are shown in
section 4.2.1). The same neuropsychological scores were included as well (Table 2.2).
Twenty-four PD (mean age: 68.38±4.73, 6 females, 18 males) and fifteen age-matched NC
subjects were recruited through the movement disorder clinic at UBC Hospital. All subjects went
through cognitive assessment with Montreal Cognitive Assessment (MoCA) and questionnaire
evaluations for depression scale, apathy, and fatigue with BECK Depression Inventory II, Apathy
Scale (SAS), the Lille Apathy Rating Scale (LARS), and Fatigue Severity Scale (FSS). In addition
to full score, sub-scores in MoCA were reported as well, which evaluated cognitive functions in
multiple domains: visuospatial and executive functions, rapid naming and lexical retrieval ability,
concentration, attention, language ability, abstraction, memory, calculation, and orientation skills
[Julayanont and Nasreddine, 2017]. PD subjects also went through clinical evaluations with MDS-
UPDRS Part I, II, III, and IV for motor and non-motor experience, motor examination, and motor
complications (Table 5.1). Both image scans and clinical evaluations were done in on medication
state. Ethics approval was issued by the University of British Columbia's Research Ethics Board
and all subjects provided signed consent forms.
PD subjects (mean±SD) NC subjects (mean±SD)
demographics age 68.38±4.73 69.4±4.76
gender 6 females/18 males 5 females/10 males
disease duration 9.92±5.86 no data
UPDRS I 9.33±5.97 no data
UPDRS II 10.92±6.51 no data
UPDRS III 27.25±8.67 no data
UPDRS IV 1.29±0.91 no data
HY 2.13±0.61 no data
cognitive & affective scores
128
MoCA full score 26.08±2.24 26.93±1.96
MoCA visuospatial/executive 4.63±0.71 4.53±0.64
MoCA picture naming 3±0 3±0
MoCA attention 5.25±0.85 5.6±0.51
MoCA language 2.38±0.64 2.33±0.62
MoCA abstraction 1.96±0.20 1.87±0.35
MoCA memory-delay recall 3±0.38 3.67±1.18
MoCA orientation 5.92±0.28 5.87±0.35
BECK depression scale 7.92±5.02 3.07±3.90
SAS 11.53±5.03 8.07±5.51
LARS -24.75±5.41 -28.8±4.04
FSS 3.93±1.53 2.36±1.32
Table 5.1 Clinical, demographical, and cognitive assessment scores in PD and NC. Bold font indicates
significant differences between two groups in t-tests with p<0.05.
[UPDRS: the Unfied Parkinson’s Disease Rating Scale, HY: Hoehn and Yahr scale, MoCA: Montreal
Cognitive Assessment, SAS: apathy scale, LARS: the Lille Apathy Rating Scale, FSS: Fatigue Severity Scale]
5.2.2 Imaging acquisition
All subjects underwent imaging studies at the University of British Columbia (UBC) MRI
Research Centre with a Philips Achieva 3.0 Tesla MRI scanner. Resting-state functional MRI
(rsfMRI) data were acquired using an 8 channel head coil and an echo-planar imaging sequence
with the following parameters: 3×3×3 mm3 resolution, 36 slices, 2000 ms TR, 30 ms TE, 90 degree
flip angle, and 240 volumes/dynamics. 3 Dimensional (3D) T1 weighted images were acquired
with 1×1×1 mm3 resolution, 60 slices, 28 ms TR, 4 ms TE and 27 degree flip angle.
5.2.3 Preprocessing
For the studies of reproducibility and MS subjects, preprocessing was the same as previous
chapters. Image preprocessing steps were performed in each subject’s native space with the
functions of slice timing and motion correction from Statistical Parametric Mapping 8 (SPM8,
129
University College London, London) for correcting temporal and spatial differences. For
registration, the FMRIB's Linear Image Registration Tool (FLIRT) from the FMRIB Software
Library 6.0 (FSL, FMRIB, Oxford) was used and a brain mask was applied to remove non-brain
areas before registration. Cortical parcellation was done on the T1-weighted images in Freesurfer
version 4.5.0 (Massachusetts General Hospital, Boston) and thirty-six cognition-associated
regions-of-interest (ROIs) were selected (Table 5.2). These ROIs have been commonly reported
in the neuropsychological literature and frequently used to investigate the relations between
cognition and resting-state functional connectivity (rsFC). Finally, the average fMRI time courses
among voxels within individual ROIs were extracted using self-programmed scripts in Matlab
(The MathWorks, Inc.) and the data were detrended before connectivity analyses.
Bilateral ROIs
superior frontal gyrus
medial frontal gyrus
inferior prefrontal cortex
temporal pole, insula, and amygdala merged
superior temporal cortex
posterior parietal cortex
post central cortex
supramarginal region
middle temporal lobe, hippocampus, hippocampal gyrus merged
occipital-parietal area
lateral occipital lobe
anterior cingulate cortex
posterior cingulate cortex
precuneus
medial orbitofrontal cortex
lateral orbitofrontal cortex
fusiform gyrus
superior parietal cortex
Table 5.2 Eighteen bilateral regions-of-interest (ROIs) in the connectivity analysis in MS study and healthy
subjects for reproducibility. Some ROIs are merged as one because these ROIs are anatomically small and
geographically close.
130
In PD and age-matched control subjects, AFNI (NIMH, Bethesda) software package was used for
fMRI preprocesses, including despiking, slice timing correction, 3D isotropic correction (3x3x3
mm3 resolution), and motion correction with rigid body alignment. Whole brain parcellation was
done in FreeSurfer 6.0 on the T1-weighted images and the structural images were then registered
to the fMRI images using rigid registration. This registration step provided the FreeSurfer
segmented ROI mask in the fMRI space. All analyses were done in the native fMRI space rather
than transforming all fMRI data to a common template, which was designed to prevent introducing
any unwanted distortions to fMRI. In the next step, several sources of variance such as head-
motion parameters, white-matter, and CSF signals were removed using linear regression. The
fMRI signals of the same ROIs were extracted and detrended to remove any linear or quadratic
trends and smoothed with 6 FWHM Gaussian kernels. Finally, bandpass filtering was performed
to retain the signal between the recommended frequencies of interest for resting state activity (0.01
Hz to 0.08Hz).
5.2.4 Dynamic functional connectivity analysis
For testing reproducibility, a sliding window with Pearson’s correlation, which is the most
common approach in the literature, was applied to calculate the windowed correlation matrices in
11 subjects with a window length of 20-time points. The window was moved 1-time point forward
in each Pearson’s r calculation, resulting in 221 correlation matrices for each subject. Seven
network features were acquired based on learned dynamic connectivity: Network Variation (NV),
Network Power (NP), Flexibility of Interhemispheric Connectivity (homologous connections,
FOCcs), Flexibility of Cross-hemispheric Connectivity (non-homologous connections, FOCcns),
Flexibility of Intrahemispheric Connectivity (within hemisphere, FOCw), Flexibility of Left-
131
hemispheric Connectivity (within left hemisphere, FOCwl), and Flexibility of Right-hemispheric
Connectivity (within right hemisphere, FOCwr). These features summarized how connectivity
patterns change between two correlation matrices (Figure 5.1, Table 5.3). Network power (NP)
measured the summed values of the dynamic functional connectivity pairs in all the windows and
the values were divided by the number of non-zero elements in each window. Network variation
(NV) calculated the differences of connectivity values between two adjacent windows and the
differences of each connectivity pairs were summed up and divided by the number of non-zero
elements in each window. Flexibility of interhemispheric connections (FOCcs) calculated the
connectivity differences of homologous connections only between two windows and then the
values were summed up to form one measure. Flexibility of cross-hemispheric connections (non-
homologous regions, FOCcns) was the measure of summed connectivity differences of non-
homologues connections between two windows. Flexibility of intrahemispheric connections
(within hemisphere, FOCw) measured the connectivity differences of connections within left and
right hemisphere between two windows and all the values were then summed up to form one value,
which can be further divided into Flexibility of right hemispheric connections (FOCwr) and
Flexibility of left hemispheric connections (FOCwl)
Repeated measure ANOVA (rmANOVA) and coefficient of variation (COV) (standard deviation
divided by the mean) were calculated to test reproducibility across time.
132
Figure 5.1 The sliding window approach with Pearson’s correlation. The global features include (A) Network
Variation (NV) and (B) Network Power (NP). The specific features include (C) Flexibility of
Interhemispheric Connectivity (x = homologous connections, FOCcs), (D) Flexibility of Cross-hemispheric
Connectivity (x = non-homologous connections, FOCcns), (E) Flexibility of Intrahemispheric Connectivity
(x = connections within hemisphere, FOCw), (F) Flexibility of Left-hemispheric Connectivity (x =
connections in left hemisphere, FOCwl), and (G) Flexibility of Right-hemispheric Connectivity (x =
connections in right hemisphere, FOCwr).
133
For the analysis in both MS and PD data sets, a sliding window approach (with a window length
of 20 time points) and the inverse covariance matrix of the ROI time courses was used to estimate
connectivity. Such a sliding window and inverse covariance matrix approach has been applied to
capture dynamic functional connectivity and accurately estimate direct connections between brain
regions in fMRI [Hutchison et al., 2013a; Smith et al., 2013]. Since the TR was 2 seconds, the 20-
point window length (WL) was 40 seconds in length, consistent with previous recommendations
[Zalesky and Breakspear, 2015]. The window was shifted one time point at a time, resulting in 221
windowed inverse covariance matrices in total for each subject. Afterwards, 6 network features
were acquired based on the learned dynamic functional connectivity. In addition to the features
mentioned in the last paragraph, another feature was calculated only based on regularized matrices.
Density (DEN) estimated how dense the connections were by taking all non-zero connectivity
values and divided by all possible connections. In short, DEN and NP measured how dense and
strong the overall connectivity was, while NV and flexibility measures calculated global and
specific network dynamics, respectively. All the connectivity values in every measure (except
DEN, as DEN focused on quality rather than quantity) were squared first, summed and then the
square root of the sum was taken. All values (the square root of the sum) across windows were
summed and then divided by the total number of windows. Therefore we did not take into account
the effects of positive and negative correlation between two ROIs in our analyses, and instead,
considered how connectivity strength changed. Although these network features were all
calculated based on learned dynamic connectivity matrices, we considered DEN and NP as
stationary network features as they did not calculate the differences between two matrices. Rather,
they represented the average values across the scanning time. Figure 5.2 demonstrates these
network features in a graph fashion. Table 5.3 describes the mathematical definitions of each
134
measure. In MS, due to the sparsity of connectivity matrices, FOCw was not divided into FOCwl
and FOCwr. Therefore, in total, there were 6 dynamic features in MS and 8 features in PD.
Figure 5.2 The sliding window approach with inverse covariance matrix. (A) Network Power (NP) calculates
the average connectivity strength across windows. (B) Density (DEN) computes how dense the existing
connections are across windows. (C) Network Variation (NV) estimates the average global connectivity
changes between two adjacent windows across time. (D) Flexibility of Interhemispheric Connections
(FOCcs), (E) Flexibility of Cross-hemispheric Connections (FOCcns), and (F) Flexibility of Intrahemispheric
Connections (FOCw) measure the average connectivity changes in interhemispheric, cross-hemispheric, and
intrahemispheric connections between two adjacent windows across time, respectively. These features are
illustrated in a graph fashion as the mathematical definitions are similar to (C) (details in Table 5.3). M ij (t)
135
Network Features Definitions
Density (DEN) ∑(𝑀𝑖𝑗−𝑛𝑛𝑧(𝑡)/𝑡𝑜𝑡𝑐𝑜𝑛)
𝐿
Network power (NP) ∑ √∑ 𝑀𝑖𝑗(𝑡)2
𝑖𝑗
𝑀𝑖𝑗−𝑛𝑛𝑧(𝑡)
𝐿
Network variation (NV) ∑ √∑ (𝑀𝑖𝑗(𝑡) − 𝑀𝑖𝑗(𝑡 − 1))2
𝑖𝑗
𝑀𝑖𝑗−𝑛𝑛𝑧(𝑡)
𝐿
Flexibility of interhemispheric connections (FOCcs) ∑ √∑ (𝑀ℎ𝑖𝑗(𝑡) − 𝑀ℎ𝑖𝑗(𝑡 − 1))2𝑖𝑗
𝐿
Flexibility of cross-hemispheric connections (FOCcns) ∑ √∑ (𝑀𝑛𝑖𝑗(𝑡) − 𝑀𝑛𝑖𝑗(𝑡 − 1))2𝑖𝑗
𝐿
Flexibility of intrahemispheric connections (FOCw) ∑ √∑ (𝑀𝑤𝑖𝑗(𝑡) − 𝑀𝑤𝑖𝑗(𝑡 − 1))2𝑖𝑗
𝐿
Table 5.3 Mathematical definitions of network features learned in the sliding window approach
[Mij(t) is the i-by-j connectivity matrix containing every element at time t, M ij-nnz(t) contains all the non-zero
connectivity values in the matrix Mij(t), totcon is the total possible connections in the matrix, which is the
number of ROI times the same number, Mij(t-1) is the i-by-j connectivity matrix containing every element at
time t-1, Mhij represents all homologous connections in matrix Mij, Mnij represents all interhemispheric
connections except homologous elements in matrix Mij, Mwij represents all intrahemispheric connections in
matrix Mij, L represents the total number of windowed correlation matrices]
is the i-by-j inverse covariance matrix containing every element at time t, M ij (t-1) is the i-by-j inverse
covariance matrix containing every element at time t-1. Mij-nnz (t) is the i-by-j inverse covariance matrix
containing non-zero values at time t. L represents the total number of windowed matrices.
136
Two sample t-tests were carried out to investigate significant differences of all network features
between patient and healthy subject groups. False-discovery rate control (FDR) was applied as a
method to correct for multiple comparisons.
5.2.5 Correlation and regression analyses
Correlations between individual dynamic features and behavioural scores were evaluated by
Pearson correlation. Only the correlations that survived for Bonferroni correction were reported.
In addition, a principal component analysis (PCA) was carried out to remove the inter-correlations
between dynamic features before linear regression. The PCA scores of components (explaining >
90% of the variance) were included in the linear regression model as predictors and age was also
included as covariance. Individual behavioural measures were included in the regression model as
response variables.
In the MS data set, composite and transformed/standardized scores were included such as the
Verbal Fluency Test (FAS), Working Memory Index (WMI), Processing Speed Index (PSI),
transformed Trail Making Test Part A and B (transformedTMTA/B), the Wisconsin Card Sorting
Test Complete Categories (WCSTCC), Multiscore Depression Inventory Total T Scores
(MDITOTT), State-Trait Anxiety Inventory: Trait (STAIT), State-Trait Anxiety Inventory: State
(STAIS), and Fatigue Severity Scale (FSS). In addition, age, education, EDSS, and disease
duration were included as well.
In the PD data set, demographical and clinical measures were included in the analysis such as
disease duration, gender, UPDRS scores of 4 parts, MoCA full scores, H&Y stage, apathy scores,
fatigue scales, and BECK depression scale scores. Moreover, the sub-scores of MoCA were further
137
included to evaluate functions in multiple cognitive domains: visuospatial and executive functions,
rapid naming and lexical retrieval ability, concentration, attention, language ability, abstraction,
memory, calculation, and orientation skills.
5.3 Results
5.3.1 Reproducibility
None of the network features showed significant differences among the three fMRI sessions (pr >
0.05 in all measures, Figure 5.3). Similarly, COV of most of the measures derived from three fMRI
sessions demonstrated similar values (Table 5.4). The results indicate that these measures are
reproducible across scans.
Figure 5.3 rmANOVA results of dynamic features across 3 fMRI sessions.
Dynamic network features 1st fMRI 2nd fMRI 3rd fMRI
NV 10.5% 12.9% 9.7%
NP 10.3% * 10.2% * 7.3% *
FOCcs 14.7% 19.1% 18.2%
138
FOCcns 10.3% 12.8% 9.6%
FOCw 10.6% 13.0% 9.9%
FOCwl 10.4% 11.9% 10.6%
FOCwr 11.4% 15.1% 10.1%
Table 5.4 COV results of dynamic features across 3 fMRI sessions.
[* not Gaussian]
5.3.2 Multiple Sclerosis
5.3.2.1 Feature comparison
MS subjects showed lower network variation (NV) and higher flexibility of interhemispheric
connections (FOCcs) then NC while controlling for a false discovery rate (FDR) (corrected p
values: 0.02 and 0.04, respectively, Figure 5.4 upper panel). Moreover, MS subjects also
demonstrated lower density (DEN) and network power (NP) with FDR correction (corrected p
values: 0.02 and 0.02, Figure 5.4 lower panel). These results indicated that MS subjects had overall
weaker connectivity, fewer connections, less dynamic overall connectivity, but more flexible
interhemispheric connections in MS subjects. In other words, there is a loss of dynamic
coordination in global connectivity and increased interhemispheric connectivity fluctuations in MS
subjects compared to NC.
139
Figure 5.4 Results of dynamic feature comparison between MS and NC. The upper panel shows the
differences between NC and MS in dynamic features. MS shows higher flexibility of interhemispheric
connectivity (corrected p = 0.02) and lower network variation (corrected p = 0.04). The lower panel illustrates
that MS presents lower network density (corrected p = 0.02) and network power (corrected p = 0.02). All p
values are controlled for false discovery rate (FDR).
5.3.2.2 Correlation and regression
Among all the correlation pairs, only disease duration showed significant correlations with
dynamic features FOCcns (r=-0.55, p<0.001), FOCcs (r=-0.56, p<0.001), and FOCw (r=-0.54,
p<0.001) in MS after corrections for multiple comparisons with 14 tests were run (Figure 5.5).
140
The first two PCA components of dynamic features in MS explained > 90% of the variance and
therefore they were included in the linear regression model. These two components represented
different aspects of dFC. The first component, explaining 79% of the variance, showed similar and
positive coefficients of all features (except DEN); therefore, this component mostly represented
global dynamics (Figure 5.6). The second component, explaining 11% of the variance, had NV
and NP loading negatively while the rest features loaded positively, and thus represented
connection density and the effects from dFC in specific connections rather than global dynamics
(Figure 5.6). Among all the demographical, clinical, and cognitive scores, only disease duration
appeared to be significantly modulated by these two components of dynamic features and age
(p<0.001). The loadings on both the 1st and 2nd components were negatively correlated with disease
duration, with significant p values of 0.00088 and 0.03705, respectively. Table 5.5 and Figure 5.7
illustrate the results of linear regression analysis.
141
Figure 5.5 Significant correlations between disease duration and dynamic features in MS. All correlation
pairs survive for Bonferroni correction.
[FOCcns: flexibility of cross-hemispheric connectivity, FOCcs: flexibility of interhemispheric connectivity,
FOCw: flexibility of intrahemispheric connectivity, DD: disease duration]
142
Figure 5.6 PCA results of dynamic features in MS. These two components explain 90% of the variance and
are included in the linear regression analysis as predictor data.
[FOCcns: flexibility of cross-hemispheric connectivity, FOCcs: flexibility of interhemispheric connectivity,
FOCw: flexibility of intrahemispheric connectivity, NV: network variation, NP: network power, DEN:
density]
For individual predictors
response/
behavioural score
predictors /PCA components
of dynamic feature
estimate in
regression model
standard
error
p
values disease duration component 1 -1.56 0.43 0.00088
component 2 -2.50 1.16 0.03705
age (covariate) 0.37 0.09 0.00015
For the whole model
Number of observations: 46, Error degrees of freedom: 43
Root Mean Squared Error: 6.25
143
R-squared: 0.523, Adjusted R-Squared 0.489
p-value = 6.81e-07
Table 5.5 Significant linear regression model in MS.
Figure 5.7 The adjusted variable plot of the linear regression model of PCA components and disease duration
in MS. The details of adjusted whole model is shown in Table 5.5. The plot shows the fitted responses
(adjusted disease duration) with the other predictors (2 components of dynamic features and age) averaged
in the model (adjusted whole model).
5.3.3 Parkinson’s Disease
5.3.3.1 Feature comparison
PD subjects did not show any significant differences of dynamic features compared to NC.
144
5.3.3.2 Correlation and regression
None of the clinical measures were significantly correlated with dynamic features. However,
FOCcs was correlated with MOCA sub-score delayed recall (r=0.54, p=0.006, Bonferroni
correction corrected with 7 tests of MoCA sub-scores were run).
Figure 5.8 Correlation between FOCcs and MoCA sub-score DELY in PD. The dynamics of
interhemispheric connections are significantly correlated with scores of memory sub-test in MoCA with
r=0.54 and p=0.0060, corrected for Bonferroni correction.
[FOCcs: flexibility of interhemispheric connections, DELY: memory test – relay recall]
The first four PCA components of dynamic features explained > 90% of the variance and they
were included in the linear regression analysis (Figure 5.9). The 1st component explained 42 % of
the variance and mostly represented overall dFC, while the 2nd component explained 29 % of the
variance, in which non-interhemispheric dynamics and interhemispheric dynamics/connectivity
145
strength loaded heavily. Hence, this component expressed interhemispheric vs non-
interhemispheric dFC. The 3rd component explained 13% of the variance and represented
interhemispheric vs non-interhemispheric dFC but with less effects from overall connectivity as
NP and DEN both showed limited loadings. The 4th component, explaining 7% of the variance,
loaded on all dynamic features equally except FOCw, which implied that this component
represented overall dynamics with less effects from the variation of intra-hemispheric
connectivity.
Figure 5.9 Four principal components of the dynamic features in PD.
[FOCcns: flexibility of cross-hemispheric connectivity, FOCcs: flexibility of interhemispheric connectivity,
FOCw: flexibility of intrahemispheric connectivity, FOCwr: flexibility of right intrahemispheric
connectivity, FOCwl: flexibility of left intrahemispheric connectivity, NV: network variation, NP: network
power, DEN: density]
In the linear regression analysis, UPDRS III and memory sub-score of MoCA were predicted by
components of dynamic features with p=0.0177 and 0.006, respectively. In the model of UPDRS
III against components of dynamic features, component 4 demonstrated significant effects
146
(p=0.001) with a negative estimate. On the other hand, in the model of MoCA memory scores
against components of dynamic features, component 3 and age both showed significant
contributions with p=0.047 and 0.02 with negative estimates, respectively (Table 5.6, Figure 5.10).
Model 1
For individual predictors
response/
behavioural score
predictors / PCA components
of dynamic feature
estimate in
regression model
standard
error
p
values UPDRS III component 1 0.80 0.92 0.39394
component 2 0.81 0.95 0.40680
component 3 -1.02 1.46 0.49399
component 4 -7.56 1.94 0.00106
age (covariate) -0.21 0.36 0.56314
For the whole model
Number of observations: 24, Error degrees of freedom: 18
Root Mean Squared Error: 6.88
R-squared: 0.507, Adjusted R-Squared 0.37
p-value = 0.0177
Model 2
For individual predictors
response/
behavioural score
predictors / PCA components
of dynamic feature
estimate in
regression model
standard
error
p
values MoCA sub-score:
Delay recall
component 1 0.02 0.14 0.89292
component 2 -0.18 0.14 0.22987
component 3 -0.47 0.22 0.04703
component 4 0.57 0.29 0.06325
age (covariate) -0.14 0.05 0.02096
For the whole model
Number of observations: 24, Error degrees of freedom: 18
Root Mean Squared Error: 1.03
R-squared: 0.569, Adjusted R-Squared 0.449
p-value = 0.00608
Table 5.6 Two significant linear regression modes in PD.
147
Figure 5.10 The adjusted variable plot of linear regression modes in PD. The plot shows the fitted responses
(adjusted UPDRS III and Delay Recall Scores) with the other predictors (4 components of dynamic features
and age) averaged in the model (adjusted whole model). Details of adjusted whole model are shown in Table
5.6.
5.4 Discussion
Our results of reproducibility demonstrated that even though dynamic functional connectivity
underlined how neural patterns change temporally, the dynamic network features were still quite
reproducible across scans in this study. As section 4.4 stated, advanced statistical methods can be
used to ensure the robustness of reproducibility, but our preliminary results justified the use of
dynamic network features to study brain organization.
5.4.1 Changes in dynamic functional connectivity may indicate compensation in MS
In this study, we observed that average connectivity strength (i.e. network power) and density were
both reduced in MS, and that MS subjects had overall reduced dynamic functional connectivity.
Interestingly, interhemispheric connectivity was more variable compared to NC, possibly to
148
compensate for overall decreased global connectivity. Our results are consistent with previous dFC
studies in MS [D’Ambrosio et al., 2018]. Previous research had concluded that MS subjects spent
less time in “high-connectivity” states, which means that networks were more often to be weakly
connected than strongly connected. This phenomenon was similar to our findings that average
connectivity strength was reduced and connectivity was less dense in MS. Moreover, fewer
switches were observed in the previous study in MS [D’Ambrosio et al., 2018], which indicates a
reduction of dynamics in MS. Similarly, we observed reduced network variation in MS, suggesting
that global connectivity did not change as much as in normal subjects across time. This finding
also supports previously-reported dynamic connectivity reduction in MS subjects. However, as
previous studies did not target specific connections, the dynamic balance between different
connections was not captured. Our results demonstrated that while global connectivity in MS
subjects lost dynamic coordination, interhemispheric connections became more dynamic, possibly
demonstrating a compensatory effect to overcome the functional disruption elsewhere and
structural damage (i.e. demyelination) in the corpus callosum [Bodini et al., 2013]. Such
compensatory effect in dFC is also supported by the results in Chapter 3. In Chapter 3, we
discovered that stronger interhemispheric connections in MS are homogenized and associated with
better cognitive performance, representing compensatory effects. Therefore, we concluded that
when structural damages impact information transfer across hemispheres via the corpus callosum,
the interhemispheric connections become stronger as well as more dynamic (i.e. fluctuate more)
in order to maintain brain function. Taken together, we propose that global rsFC in MS is weaker,
sparser, and more rigid; on the other hand, other specific connections such as interhemispheric
connections may be stronger and more dynamic to compensate for functional disruption elsewhere
and structural damage in callosal connections.
149
In PD, we did not observe any differences of dynamic features between PD and NC, which is
possibly due to the relatively mild disease severity in this cohort. However, the index of overall
disease severity (i.e. UPDRS III score) was related to dynamic features, which is discussed in the
following section.
5.4.2 Disease duration is robustly modulated by dynamic functional features in MS
In both correlation and regression analyses, only disease duration exhibited association with
dynamic connectivity features. Longer disease duration was associated with decreased dynamics
in interhemispheric, intra-hemispheric, and cross-hemispheric connectivity, implying that as
disease progressed, functional connectivity lost dynamic coordination of distributed brain regions
connecting the two hemispheres and regions within the same hemisphere. In the linear regression
model, different aspects of dynamic patterns were affected by disease duration. The loadings on
both components were negatively correlated with disease duration, supporting the conclusion that
disease progression is associated with the loss of dynamic coordination across the brain in MS.
Previous studies of dFC in MS subjects have conjectured that decreased dynamics served as
biomarkers [D’Ambrosio et al., 2018; Leonardi et al., 2013], yet the relations between dFC and
clinical data remain unclear; similarly, whether clinical presentation and cognitive function can be
predicted by dFC requires further investigation. In this study, we partially addressed this question
by showing that disease duration is significantly associated with dynamic features in a linear
regression model. Although advanced statistical methods are required to predict disease
progression and any other behavioural outcomes, our findings shed light on exploring dynamic
connectivity and disease effects.
150
Interestingly, EDSS did not show any significant relations to dynamic features but disease duration
did. Based on the current literature, we assume that dFC is more related to cognitive function rather
than motor and physical ability. As EDSS mainly evaluates disease severity based on physical
disabilities, it is possibly that this disease index is less related to dFC. Rather, a more general index,
such as disease duration, contains information of cognitive decline along disease course and
exhibits relations to dFC.
We did not observe any associations between cognitive performance and dynamic features in the
correlation analysis. Similarly, none of the cognitive scores was modulated by the components of
dynamic features. However, when more PCA components were included in a linear regression
model (i.e. data explained almost 100% of the variance), component 4 appeared to be significantly
related to the Wisconsin Card Sorting Test Complete Categories (WCSTCC) (p=0.011,
Appendices D.1). In this cognitive task, the ability to switch strategies based on rules is required
(i.e. cognitive flexibility), which has been shown to be related to dynamic functional connectivity
[Douw et al., 2016]. Yet this component only explained 1.8% of the variance and flexibility of
interhemispheric connections (FOCcs) loaded positively and heavily onto this component
(Appendices D.2). Therefore, this regression pattern implied that the dynamic fashion of
interhemispheric connections might support the performance of strategy switching, but such
pattern may not be representative enough in our data as the component explained a limited amount
of variance.
We conclude that drsFC can be indicative of disease progression, which could serve as a
biomarker, and dynamic coordination of homologous regions might be supportive of strategy
switching ability, but such brain-behavioural association is very mild in our cohort.
151
5.4.3 Disease severity can be predicted by dynamic features in PD
In the linear regression model, the UPDRS III score could be significantly predicted by the 4th
component of dynamic features. This component only explained 7% of the total variance in
dynamic features and represented overall dynamics with fewer effects from intra-hemispheric
connectivity. As this component was negatively weighted in the regression model, it was
interpreted that higher overall dynamics was associated with lower UPDRS III scores, implying
that higher disease severity was related to reduced dFC in PD. Previous research has shown that
weaker connections were more dynamic but stronger connections lost dynamic coordination in PD
[Díez-Cirarda et al., 2018; Kim et al., 2017]. As we did not observe differences in dynamic
features, it is hard to conclude whether such dynamic balance between strong and weak
connections exist in our data. However, we do show that higher disease severity is related to
decreased dFC in PD, which is partially consistent with previous studies. One possible explanation
for this is that different sub-types of motor symptoms in PD have been linked to different types of
cognitive deficits [Kehagia et al., 2012; Miller et al., 2013; Thenganatt and Jankovic, 2014].
Tremor-dominant PD subjects often demonstrate disabilities in planning, working memory, and
executive functions; therefore, this subtype has been associated with the cognitive “frontal
subtype” and dopamine depletion. The other subtype is associated with pronounced gait
disturbance and often shows deficits in visuospatial function and sematic fluency, which has been
related to the cognitive “posterior subtype” and cholinergic deficits in PD. Therefore, although
UPDRS III is assessed based on motor symptoms, dopamine depletion not only impairs motor
loops but also frontal-striatal circuits which are important for many cognitive functions [Kehagia
152
et al., 2012]. Therefore, the relations between UPDRS score and dynamic features in this study
may be an indirect indicator of the relations between frontal subtype and dFC in PD.
5.4.4 Memory function is robustly associated with dynamics in PD
The MoCA sub-score of memory delayed recall screens deficits in retrieval memory, and was the
only MoCA sub-sore that demonstrated associations in both correlation and linear regression
analyses. With correlation analysis, FOCcs was positively correlated with scores of delayed recall,
suggesting that higher dynamics of interhemispheric connections were related to better retrieval
memory function in PD. In the linear regression model, delayed recall performance was modulated
by component 3 of dynamic features and age. Component 3 explained 13 % of the variance and
mainly represented interhemispheric dFC with negative loadings and intra-hemispheric dFC with
positive loadings as other features showed limited loadings. Because component 3 showed a
negative regression coefficient, higher FOCcs and lower intra-hemispheric dFC predicted better
delayed recall performance as they loaded positively and negatively, respectively. Therefore, taken
together, both correlation and linear regression suggest that better retrieval memory was not only
correlated with, but also could be predicted by, higher dynamics in interhemispheric connectivity
in PD, strengthening the important role of dynamic coordination between two hemispheres on
cognitive process.
Delayed recall scores were also modulated by age, which acted as covariate in the regression
model. As age had a negative regression coefficient, the model also reinforced the links between
aging and decreased cognitive ability in the memory domain. Finally, we also propose that memory
function might be strongly related to dynamics of interhemispheric connections, but it might be
153
more influenced by aging than dynamic features when they were jointly assessed together, given
that component 3 was almost not significant in the linear regression model.
In this study, we did not evaluate cognitive functions in PD subjects with a comprehensive
neuropsychological battery other than MoCA. Therefore, although we only observed relations
between memory function and dFC, it did not mean other cognitive domains, especially executive
function, are unrelated to dFC. In fact, as the sub-tests in MoCA which evaluate attention/executive
function are relatively easy for subjects without dementia or MCI, using only MoCA might be
insensitive to mild executive dysfunction in the early stage. Further test batteries which specifically
evaluate set-shifting, mental calculation, and other executive skills are needed to explore brain-
behaviour associations in PD, which may provide further information to understand the disease
and develop cognitive training as a treatment for cognitive impairments.
5.4.5 Limitations
Choosing ROIs for connectivity analysis is always challenging. As the focus of the study is to
explore the relations between dFC and cognition in neurological disorders, we selected the
cognition-related ROIs that are commonly reported in the literature as described in previous
chapters. Furthermore, in order to reduce the computational demand of calculating inverse
covariance matrix in the sliding window approach, only a relatively fewer ROIs associated with
cognition were included in the analysis. For future research, more regions should be included
regardless whether they are related to cognition or not from a traditional point of view.
154
5.5 Conclusion
In this chapter, the goal was to investigate 1) whether dFC features can serve as biomarkers for
neurological diseases and 2) how dFC features are related to cognitive functions in PD and MS –
two distinct diseases with similar cognitive impairments such as executive dysfunction. In MS,
average connection density, strength, and global dynamics were all decreased, but
interhemispheric connections became more variable, possibly compensating for structural damage
in the corpus callosum and other regions. In PD, we did not observe differences of any dynamic
features compared to control subjects, which might be due to the relatively mild disease severity
in our cohort. In the MS subjects, the brain-behaviour analyses revealed that higher disease
duration was associated with decreased dFC and such association could be predicted by the
principal components of dynamic features. Set-shifting abilities measured in WCST were
predicted by the component representing dFC in interhemispheric connections; however, this
relation did not explain a significant proportion of the variance in the functional connectivity data.
In PD, better performance of memory tests was robustly related to/predicted by higher dynamics
in interhemispheric connectivity, while overall disease severity was modulated by global dynamics
with fewer effects from intra-hemispheric connectivity. Taken together, these brain-behaviour
associations provide insights into understanding the disease effects in different neurological
conditions.
155
Chapter 6: Dynamic brain organization, disease severity, and executive
function
In this chapter, different aspects of functional connectivity are integrated to study the relations
between stationary, dynamic functional connectivity, cognitive functions, disease severity, and
demographical factors. By combining stationary and dynamic functional connectivity, dynamic
brain organization can be assessed, which provides a more comprehensive approach to investigate
the relations between functional connectivity, disease progression, and cognitive function as well
as compensatory effects.
6.1 Introduction
6.1.1 Stationary and dynamic functional connectivity
Functional connectivity in human brain networks is usually estimated by calculating the temporal
correlation (e.g. Pearson’s r correlation) or neural influence (estimated by e.g. conditional
dependence) between brain signals among anatomically-segregated brain regions [Friston, 1994;
Friston, 2011]. These methods usually assume that the estimated functional connectivity is
temporally stationary, i.e. does not change over time. However, connectivity fluctuates across time
from seconds to minutes even in the resting state, which can be estimated by models of dynamic
functional connectivity [Allen et al., 2014; Betzel et al., 2016; Chang and Glover, 2010;
Handwerker et al., 2012; Hutchison et al., 2013a; Jones et al., 2012]. The simplest, and perhaps
most common time-varying approach to assess dynamic functional connectivity is to estimate
156
correlations between brain regions within a fixed-length, sliding window, with the (possibly
overlapping) windows ultimately moved over the entire data. Details regarding this method has
been discussed in sections 1.2.2 and 5.1.1. The time-varying approach can capture the temporal
variability of functional connectivity, but the spatial network characteristics cannot be estimated
with this method. A graph theoretical approach appears to be a powerful technique to obtain
topological information of networks, which quantifies information flow in stationary functional
connectivity globally and locally.
6.1.1.1 Cognition and stationary functional connectivity: graph theoretical analysis
Graph theoretical analysis summarizes network characteristics and can represent how efficiently
the networks propagate information globally and locally [Sporns, 2013]. Global efficiency, a
measure of integration, has been related to working memory and verbal comprehension; while
modularity, a measure of segregation, correlates with motor task performance [Cohen and
D’Esposito, 2016; Pamplona et al., 2015]. Hub structures are related to higher-order cognitive
functions such as executive function [Reineberg and Banich, 2016]. With graphical metrics, the
brain can be divided into modules, with each module related to an individual cognitive component
[Bertolero et al., 2015]. These “cognitive brain modules” are linked by hub structures and the hubs
with higher connectivity engage more cognitive modules. Alterations in graphic theoretical
measures of the computed brain stationary connectivity networks have been associated with a
variety of cognitive abilities and disease populations. Altered hub structure has been demonstrated
in people with Parkinson’s disease with attention/executive deficits, with hub nodes having
reduced importance [Baggio et al., 2014]. Global network measures cannot significantly
distinguish depressed subjects from healthy controls; however, with a support vector machine
157
approach, individuals have been accurately classified with network measures [Lord et al., 2012],
suggesting that a combination of machine learning methods and connectivity features is beneficial
to assist diagnosis.
6.1.1.2 Cognition and dynamic functional connectivity: time-varying approach
Dynamic functional connectivity, as also referred to network dynamics and assessed by time-
varying approaches, appears to be particularly pertinent to several cognitive processes including
memory, language, attention, and executive functions [Braun et al., 2015; Bressler and Scott Kelso,
2001; Kucyi et al., 2016; Mattar et al., 2015; McIntosh et al., 2008; Shafto and Tyler, 2014;
Thompson et al., 2013]. Increased dynamical variability in the EEG was found to be correlated
with better performance (i.e. shorter reaction time and higher accuracy) in a memory task,
emphasizing the importance of brain complexity in cognitive development [McIntosh et al., 2008].
There are associations between dynamic changes in frontoparietal/frontotemporal networks and
neuropsychological measures, showing that the flexibility of neuronal activity in the frontal
regions is cognitively beneficial for working memory performance and executive functioning
[Braun et al., 2015]. This has led to a proposed “functional cartography” of the cognitive system,
based on the estimated amount of integration and recruitment of brain regions during different
cognitive processes [Mattar et al., 2015]. Network dynamics may also be important for language
function in an aging population [Chai et al., 2016; Shafto and Tyler, 2014]. Measures of dynamic
functional connectivity have been recently applied to understand how the human brain is affected
by diseases such as Parkinson’s disease, Schizophrenia, Alzheimer's disease, and major depression
[Damaraju et al., 2014; Kaiser et al., 2015; Madhyastha et al., 2014; Sakoglu et al., 2010; Wee et
al., 2013].
158
6.1.2 Studies looking at both stationary and dynamic aspects of functional connectivity
Although investigating graph theoretical metrics and dynamic connectivity have become
increasingly important in clinical and neuroscience (please refer to section 4.1.2, 4.1.3, and 5.1.3
for review), not many studies evaluate both aspects simultaneously. A few studies have performed
graph theoretical analysis on the windowed matrices derived from the sliding window approach to
investigate how brain topological features change across time and the variation of global network
characteristics [Chen et al., 2016a; Chiang et al., 2016; Kim et al., 2017; Zalesky et al., 2014]. This
approach has been referred to as dynamic graph analysis and the most commonly used metrics are
efficiency and modularity [Preti et al., 2017], which offers a promising way to study the balance
between integration and segregation.
Compared to other resting-state networks (RSNs) such as sensory-motor, auditory, visual
networks, the nodes in the salience network (SN), subcortical regions, and the frontoparietal
network (FPN) showed higher flexibility and diversity (i.e. spatially varied), suggesting that the
SN, FPN, and subcortical regions are more dynamic and interact with the nodes outside of the
community determined by graphical analysis. Moreover, the spatiotemporal brain organization of
the SN predicted processing speed, attention, executive functioning, and cognitive flexibility,
suggesting that the SN is a hub to facilitate cognitive function by interacting flexibly with other
networks. Furthermore, the dynamics of topological characteristics have been profiled. The hub
and inter-modular connections have shown to be more dynamic, which are mainly located in the
FPN and the default mode network (DMN) and included in long anatomical connections; while
non-hub regions such as the cerebellum, vermis, and some nodes in temporal areas connectivity
appeared to fluctuate less [Zalesky et al., 2014]. Of note, the FPN, DMN, and long anatomical
159
connections are highly related to cognitive function (detailed review can be found in section 1.1
and 1.4). Such brain dynamics possibly react to the balance between different principles of brain
organization (i.e. efficiency of information transfer, process, and energy cost) as well as cognitive
processes.
This approach for studying dynamic brain organization has been applied to a few clinical studies.
In patients with epilepsy, the network characteristics represent local segregation was more
dynamic; while the measures represent global integration was more stationary [Chiang et al.,
2016]. In Parkinson’s disease (PD), global efficiency showed higher variation across time
windows, suggesting that the parallel information transfer was less efficient compared to that in
healthy subjects [Kim et al., 2017]. Taken together, dynamic brain organization provides insights
into how diseases affect brain connectivity spatially and temporally, which potentially helps
predict disease severity. However, none of the clinical research has related dynamic brain
organization to cognitive performance.
6.1.3 Study aims
In this study, the overarching purpose was to study how dynamic and static functional connectivity
related to cognitive performance in MS and PD. We applied a sliding window approach and graph
theory analysis to assess dynamic functional connectivity and stationary network characteristics in
both neurological conditions, aiming to investigate the utility of these measures. In addition, a
machine learning method, multiset canonical correlation analysis (MCCA), was applied to explore
the associations between dynamic and stationary functional connectivity features, and behavioural
data in order to explore models that may be beneficial for treatment development. Since previous
neuroimaging studies have shown that both dynamic functional connectivity and graphical
160
measures are beneficial to strengthen understanding of complex cognitive processes, we
hypothesize that subjects with neurological disorders demonstrating cognitive dysfunction would
also demonstrate abnormal dynamic and stationary functional connectivity. Furthermore, given
that human cognition is complex and operates as a network phenomenon [Petersen and Sporns,
2015], we also hypothesize that cognitive functions in neurological disorders will be jointly
affected by dynamic and stationary connectivity networks as well as disease severity.
6.2 Materials and methods
6.2.1 Subjects
PD data are from GFM2 project and MS data were from COGMS project. Both cohorts are the
same as chapter 5.
Ethics approval was issued by the University of British Columbia's Clinical Research Ethics Board
and all subjects provided written, informed consent.
Forty-six Relapsing-Remitting Multiple Sclerosis (RRMS) patients were included in the study and
all the subjects underwent both cognitive testing and Magnetic Resonance imaging (MRI).
Demographics are shown in Table 2.2. All patients fulfilled the McDonald 2005 criteria [Polman
et al., 2005] for diagnosis of MS and were recruited from the MS clinic at the University of British
Columbia Hospital. Exclusion criteria included the following: 1) subjects with significant
depression and/or other psychiatric illness, 2) history of drug or alcohol abuse, or 3) use of steroids
in the last 3 months.
161
The same twenty-four PD patients as the last chapter were included in the study. Clinical
evaluations and demographics are shown in Table 5.1. All PD patients were in on-medication state
and severe depressed subjects were excluded in the study.
6.2.2 Neuropsychological and clinical assessments
All MS patients underwent a test battery which included Digit Span, Arithmetic, Letter-Number
Sequencing, Symbol Search, and Symbol Coding subtests from the Wechsler Adult Intelligence
Scale IV (WAIS-IV), the Verbal Letter Fluency Test (FAS), Wisconsin Card Sorting Test
(WCST), and Trail Making Test A and B (TMT A and B). The Working Memory Index (WMI) is
a composite score of Digit Span, Arithmetic, and Letter-Number Sequencing; while the Processing
Speed Index (PSI) is another composite score of Symbol Search and Symbol Coding. These two
composite scores were included in the analysis rather than the individual scores in WAIS-IV.
Clinical questionnaires were administered, which included Multiscore Depression Inventory
(MDI), State-Trait Anxiety Inventory (STAI), and Fatigue Severity Scale (FSS). The
neuropsychological battery aimed to evaluate executive skills including mental flexibility, concept
formation, attentional switching, spontaneous generation of verbal information, and working
memory as well as processing speed abilities including attention and visual scanning. The subject’s
disability severity was rated by the Kurtzke Expanded Disability Status Scale (EDSS)[Kurtzke,
1983] as scored by a neurologist at the time of recruitment.
All PD patients were assessed with motor and non-motor symptoms with Movement Disorder
Society Unified Parkinson's Disease Rating Scale (MDS UPDRS). Moreover, depression, apathy,
and fatigue symptoms were evaluated with questionnaires. Cognitive abilities were assessed with
Montreal Cognitive Assessment (MoCA). Details have been listed in section 5.2.1.
162
6.2.3 Imaging acquisition and processing
All subjects underwent imaging studies at the University of British Columbia (UBC) MRI
Research Centre with a Philips Achieva 3.0 Tesla MRI scanner. Resting-state functional MRI
(rsfMRI) data were acquired using an 8 channel head coil and an echo-planar imaging sequence
with the following parameters: 3×3×3 mm3 resolution, 36 slices, 2000 ms TR, 30 ms TE, 90 degree
flip angle, and 240 volumes/dynamics. 3 Dimensional (3D) T1 weighted images were acquired
with 1×1×1 mm3 resolution, 60 slices, 28 ms TR, 4 ms TE and 27 degree flip angle.
The preprocessing was the same as chapter 5.
In MS, image preprocessing steps were performed in each subject’s native space with the functions
of slice timing and motion correction from Statistical Parametric Mapping 8 (SPM8, University
College London, London) for correcting temporal and spatial differences. For registration, the
FMRIB's Linear Image Registration Tool (FLIRT) from the FMRIB Software Library 6.0 (FSL,
FMRIB, Oxford) was used and a brain mask was applied to remove non-brain areas before
registration. Cortical parcellation was done on the T1-weighted images in Freesurfer version 4.5.0
(Massachusetts General Hospital, Boston) and 36 cognition-associated regions-of-interest (ROIs)
were selected (Table 5.2). These ROIs have been commonly reported in the neuropsychological
literature and frequently used to investigate the relations between cognition and resting-state
functional connectivity (rsFC). Finally, the average fMRI time courses among voxels within
individual ROIs were extracted using self-programmed scripts in Matlab (The MathWorks, Inc.)
and the data were detrended before connectivity analyses.
163
In PD, similar processing steps have been applied but with AFNI (NIMH, Bethesda), including
despiking, slice timing correction, 3D isotropic correction (3x3x3 mm3 resolution), and motion
correction. Whole brain parcellation was done in FreeSurfer 6.0 on the T1-weighted images and
the structural images were then registered to the fMRI images using rigid registration. All analyses
were done in the native fMRI space rather than transforming all fMRI data to a common template.
Moreover, several sources of variance such as head-motion parameters, white-matter, and CSF
signals were removed using linear regression. The fMRI signals of the same 36 cognition-
associated ROIs were extracted and detrended to remove any linear or quadratic trends and
smoothed with 6 FWHM Gaussian kernels. Finally, bandpass filtering was performed to retain the
signal between the recommended frequencies of interest for resting state activity (0.01 Hz to
0.08Hz).
6.2.4 Connectivity analysis
Dynamic functional connectivity (dFC) was evaluated in a sliding window approach with inverse
covariance matrices and several dynamic features were calculated. Analysis details can be found
in section 5.2.4. In total, 6 features were generated including network variation (NV), flexibility
of interhemispheric connections (FOCcs), flexibility of cross-hemispheric connections (FOCcns),
flexibility of intrahemispheric connections (FOCw), density (Den), and network power (NP).
Although these network features were all calculated based on learned dynamic connectivity
matrices, we considered DEN and NP as stationary network features as they did not calculate the
differences between two matrices. Rather, they represented the average values across the scanning
time.
164
For stationary functional connectivity (sFC), graphical measures were estimated with the Brain
Connectivity Toolbox. Analysis details can be found in section 4.2.4. In total, global efficiency,
transitivity, modularity, assortativity, characteristic path length, and rich club coefficient were
computed. Note that in order to avoid having too many features and overfitting the data in PD, rich
club coefficient was not included as the highest level was only 7 in our PD data.
6.2.5 Brain-behaviour associations
We utilized MCCA to explore the associations between modalities/data sets. MCCA determines
the linear combinations of data sets that maximize the correlations between data sets. MCCA is a
popular method for blind source separation and has been used to explore the associations between
neuroimaging and clinical/behavioural data [Chen et al., 2016b; Perry et al., 2017; Sui et al., 2013].
In this study, we included five data sets to investigate the relations between dynamic rsFC,
stationary rsFC, demographics, cognitive scores, and affective variables in RRMS. Dynamic rsFC
included NV, FOCcs, FOCcns, and FOCw. Stationary rsFC included DEN, NP, global efficiency,
assortativity, characteristic path length, modularity, rich club coefficient, and transitivity. The
demographical set included education, EDSS, and disease duration. Cognitive scores included
WMI, PSI, FAS, WCST number of categories completes, TMT A and B, in which WMI, PSI, and
FAS were standardized/adjusted scores. Affective variables contained MDI, trait anxiety (STAIT),
state anxiety (STAIS), and FSS, in which MDI (total score), STAIT, and STAIS were
transformed/standardized scores. Subjects who showed more than 1 variable that was
bigger/smaller than 2 standard deviations of the mean were considered outliers. In the end, nine
subjects/outliers were excluded in the MCCA analysis, resulting in 37 MS subjects in total. Age
effects were regressed out in a linear regression model and all the data were whitened before
165
performing MCCA. A permutation test with 1000 permutations was performed to assess the
significance level of each MCCA component. In order to ensure the robustness of results, we used
a leave-one-out cross-validation. Finally, the MCCA loadings (i.e. correlation between
transformed canonical data and input scores) and standard errors of each variable were reported.
In PD, 4 data sets were included. First, dynamic rsFC included NV, FOCcs, FOCcns, and FOCw.
Stationary rsFC set included DEN, NP, global efficiency, assortativity, characteristic path length,
modularity, and transitivity. In the clinical set, scores of UPDRS III, Beck Depression Inventory
(BECK), and Apathy Scale (SAS) were incorporated. Finally, scores of MoCA sub-tests and the
final full scores were involved in the cognitive set. Due to that naming test did not show
impairments in any patients (everyone got full score), this sub-score was not included in the
cognitive set. As the sample size of PD was relatively small, we only included the most important
variables that we were interested in to avoid overfitting the data. Some demographical and clinical
variables were not contained in the analysis such as gender, disease duration, MDS UPDRS part
I, II, and IV, etc. No outliers have been detected in PD; therefore, 24 PD patients were counted in
the MCCA analysis. MCCA was performed using the same parameters and approach as previously
described.
6.3 Results
6.3.1 MCCA in MS
MCCA identified one significant component that showed moderate to strong correlations between
the linear combination of almost all data sets (p=0.01 in a permutation test, Figure 6.1). This
component represented a linear combination of features in all sets that maximally correlate with
166
each other. Dynamic features showed strong and moderate correlation to stationary features
(r=0.80) and cognitive scores (r=0.34) as well as demographics (r=0.22). Stationary features also
demonstrated associations to cognitive scores (r=0.24) but showed limited relations to other
behavioural measures (i.e. demographics and affect). Demographics were strongly associated with
affective variables (r=0.60) as well as cognition (r=0.46) and dynamic features (r=0.22). Affective
(mood) variables were not strongly linked to resting-state functional connectivity (rsFC) features.
However, affect was associated with cognitive performance tests and demographic features.
Overall, cognition was more correlated to demographics and dynamic rsFC than static rsFC and
affect. Affective variables mainly had strong relations to demographics, but they also showed mild
associations to cognition.
167
Figure 6.2 demonstrates the MCCA loadings of all variables in the significant component (p=0.01
in a permutation test). Within each set, red stars indicated the variables that showed significant
loadings and the results were mainly interpreted based on these variables. All the variables which
showed positive loadings were positively correlated with each other; likewise, all the variables that
demonstrated negative loadings were positively associated with each other. In short, higher
dynamics, stronger static connectivity strength, longer education, and better cognitive
performances were correlated with each other. On the other hand, denser static connectivity, higher
EDSS and longer disease duration, worse TMTB performance, and depression/anxiety/fatigue
were positively associated with each other.
Better executive functions on cognitive testing were related to higher dynamics, stronger stationary
connectivity, and higher education; while worse mental flexibility measured by TMT B test are
expressed with the associations of denser static connectivity, higher disease severity, and disease
comorbidity (i.e. depression, anxiety, and fatigue).
Figure 6.1 Correlation between MCCA sets in MS. The lower triangle shows the correlation coefficients
between all data sets in the significant MCCA component (p = 0.01 in a permutation test). Each column/row
represents the linear combination of the MCCA loadings from all variables within each set. The exact
correlation coefficient between two sets is illustrated in each corresponding element.
168
Figure 6.2 MCCA loadings in MS. MCCA loadings of all variables in all sets are shown with error bars
indicating standard errors. Red stars highlight the individual variables that demonstrate significant loadings
169
(p<0.05). All the variables that show positive loadings are positively associated with each other. All the
variables that present negative loadings are positively correlated with each other.
[NV: network variation, FOCcs: flexibility of interhemispheric connections, FOCcns: flexibility of cross-
hemispheric connections, FOCw: flexibility of intrahemispheric connections, DEN: density, NP: network
power, GE: global efficiency, assor: assortativity, chapath: characteristic path length, modu: modularity,
richclub: rich club coefficient, trans: transitivity, ED: education, EDSS: Expanded Disability Status Scale,
DD: disease duration, WMI: Working Memory Index, PSI: Processing Speed Index, FAS: Verbal Fluency
Test, WCSTCC: Wisconsin Card Sorting Test number of categories completes, TMTA: Trail Making Test
A, TMTB: Trail Making Test B, MDITOTT: Multiscore Depression Inventory Total T Score, STAIS:
Anxiety Inventory-State, STAIT: Anxiety Inventory-Transit, FSS: Fatigue Severity Scale]
6.3.2 MCCA in PD
MCCA also revealed one significant component that showed moderate to strong correlations
between the linear combination of all data sets in PD (p=0.02 in a permutation test, Figure 6.3).
Stationary FC and clinical data sets demonstrated the strongest correlation with r=0.82. The rest
sets all showed moderate correlations: r=0.58 for clinical set vs cognitive set, r=0.47 for dynamic
FC vs cognitive set, r=0.44 for dynamic FC vs clinical set, r=0.36 for dynamic vs stationary FC,
and r=0.31 for stationary FC vs cognitive set. In short, the linear combination of stationary features
was strongly correlated with clinical data. On the other hand, dynamic features were more
correlated with cognitive function than other sets. Clinical data and cognitive function also showed
strong correlation.
170
Figure 6.3 Correlation coefficients between 4 MCCA data sets in PD. The lower triangle shows the
correlation coefficients between all data sets in the significant MCCA component (p = 0.02 in a permutation
test). Each column/row represents the linear combination of the MCCA loadings from all variables within
each set. The exact correlation coefficient between two sets is illustrated in each corresponding element.
The MCCA loadings of the significant component are shown in Figure 6.4 (p=0.02 in a
permutation test). Within each set, black stars indicated the variables that showed significant
loadings and the error bars showed standard errors across subjects. In the dynamic FC set, NV and
FOCw showed significant positive loadings, while FOCcs demonstrated negative loadings.
Assortativity and transitivity in the stationary set both exhibited significant negative loadings. All
variables in the clinical data showed significant positive loadings. In the cognitive set (i.e. MoCA
171
scores), LAN was loaded positively and DELAY was loaded negatively. All the variables which
showed positively loadings were correlated with each other and variables with negative loadings
were positively related to each other, which resulted in a “positive-negative pattern”. In other
words, higher dynamics in global and intrahemispheric FC, higher disease severity, more severe
depression and apathy symptoms were correlated with better language ability. Higher
interhemispheric dFC, assortativity, and transitivity were related to better memory function
measured by the delayed recall test in MoCA.
Figure 6.4 MCCA loadings in PD. MCCA loadings of all variables in all sets are shown with error bars
indicating standard errors. Black stars highlight the individual variables that demonstrate significant
loadings (p<0.05).
172
[NV: network variation, FOCcs: flexibility of interhemispheric connections, FOCcns: flexibility of cross-
hemispheric connections, FOCw: flexibility of intrahemispheric connections, DEN: density, NP: network
power, assor: assortativity, chapath: characteristic path length, GE: global efficiency, modu: modularity,
trans: transitivity, UPDRS3: Unified Parkinson's disease rating scale part III, BECK: Beck Depression
Inventory, Apathy: apathy scale, VSEXE: visuospatial/executive function, ATT: attention, LAN: language,
ABS: abstraction, DELAY: memory test delayed recall, ORIEN: orientation, total score: MoCA full score]
6.4 Discussion
6.4.1 Relations between dynamic, stationary functional connectivity, education, and
cognition in MS
We found two major patterns of the associations between rsFC and behavioural data with the
MCCA approach in MS. Better working memory, processing speed, and verbal fluency abilities
were associated with higher education, stronger stationary rsFC as well as higher network
variability. The other pattern demonstrated that poor switching ability was associated with higher
depression, state and trait anxiety, fatigue, higher disease severity, longer disease duration, and
denser but perhaps more rigid rsFC (as density was anti-correlated with all dynamic features).
Previous research has proposed that more variability in functional connectivity is related to better
cognitive performance [Kucyi et al., 2016; Mattar et al., 2015; McIntosh et al., 2008]. In our
results, we also observed positive correlations between variable network features and cognitive
performance. These cognitive performances not only required attention but also a variety of
executive skills such as working memory (to temporary hold and manipulate information to
formulate an answer), processing speed abilities (to perform focused attention, visual scanning,
173
and discriminating visual details under timed conditions), and fluency (to spontaneously generate
information according to rules and retrieve information from memory) [Kreutzer et al., 2011],
which were all higher-order cognitive functions and required more neuronal resources [Miller and
Wallis, 2009].
We found higher education and stronger static connectivity strength were also associated with
better higher-order cognitive functions, in keeping with prior results [Alosaimi et al., 2017] and
“cognitive reserve theory” which suggests that education may provide neuroprotection effects and
enhance plasticity and flexibility in a variety of neural circuits imparting protection in
neurodegenerative disease [Stern, 2002; Vance et al., 2010]. Prior studies relating functional
connectivity with cognition have been variable; stronger [Smith, 2015], or weaker but more
efficient connections [Santarnecchi et al., 2014] have both been proposed for supporting cognitive
functions. We suggest, that, the balance between dynamic and stationary connectivity and
coordination of information flow might be the key factor to facilitate cognition.
6.4.2 Disease effects on cognition and comorbidity in MS
The MCCA model also suggested a relation between disease severity, comorbidity, and worsening
executive functions. Depression and anxiety are frequent comorbidities in MS and are often
associated with more severe physical disability and disease progression, and poorer quality of life
[Marrie, 2016]. Although comorbidities in MS have been extensively studied in the clinical
literature (e.g., [Alosaimi et al., 2017; Marrie, 2017; Rahn et al., 2012]), the relations between
disease severity, comorbidity, cognition, and rsFC have not been previously investigated. We
found affective variables were highly correlated with demographics as shown in Figure 6.1. Within
this association, higher disease severity, reflected by higher EDSS and longer disease duration,
174
was related to higher depression, state and transit anxiety, and fatigue. Worse TMTB performance
(higher scores) was correlated with above-mentioned associations and higher network density,
which might be more rigid as DEN was anti-correlated with all dynamic features. TMTB
performance required a great deal of executive components especially task switching abilities and
mental speed; therefore, this MCCA pattern indicated a co-existence of disease severity,
comorbidity, reduced shifting abilities, mental speed, and denser but rigid static rsFC in MS.
Recent studies have investigated the relations between cognitive decline and psychiatric
comorbidity in neurological diseases and three theories have been proposed, 1) comorbidity is
independent of cognitive decline, 2) cognitive dysfunction is qualitatively influenced by
comorbidity, and 3) comorbidity manifests the existing cognitive impairments in neurological
diseases (i.e. cognition is quantitatively affected) [Barone and Santangelo, 2010; Karadayi et al.,
2014; Mavandadi et al., 2009]. We found that disease severity exhibited the strongest association
with depression, anxiety, and fatigue; while it demonstrated the second strongest correlation to
cognitive scores. In addition, the correlation between affect and cognition was relatively mild.
Affective status might have relatively weaker interactions with rsFC but indirectly impact higher-
order cognitive functions in early-stage MS. Therefore, our results are most supportive of the
theory that psychiatric comorbidity does not directly impact cognition, but rather, it exacerbates
existing cognitive problems in MS.
6.4.3 Better memory function in PD is related to dynamic interhemispheric connectivity
and stationary network segregation
Delayed recall test in MoCA reflects memory retrieval and encoding function [Julayanont and
Nasreddine, 2017]. In PD, whether memory deficits are caused by impairments in retrieval or
175
consolidation (i.e. whether patients are unable to re-access information or acquire new
information) is still under debate [Chiaravalloti et al., 2014; Costa et al., 2014]. However, the
striatum has been proposed to primarily support memory retrieval and the deficits might be related
to executive dysfunction [Costa et al., 2014; Scimeca and Badre, 2012]. In this study, although we
were unable to pinpoint which kinds of memory deficits were represented in our data with only
sub-tests in MoCA, we demonstrated inter-correlations between better memory function, network
segregation, and dynamic coordination between two hemispheres.
In our results, better performance of memory test was associated with 1) higher transitivity and
assortativity as well as 2) higher dFC in interhemispheric connections. As transitivity is a measure
of functional segregation, higher value represents that the brain networks are more segregated than
integrated. Assortativity is a correlation coefficient of node degrees of connected pairs,
representing the tendency that similar nodes (with similar degree) are linked to each other.
Therefore, higher value represents higher resilience. Of note, the values of assortativity in all
subjects were small, which means that resilience may not be very obvious, but the subtle changes
might still be related to cognition in PD. Higher FOCcs represents higher variation in the
homologous connections across time, that is, higher dFC in interhemispheric connections. Taken
together, better memory function in PD was correlated with the segregated and resilience in the
stationary FC and also related to more dynamic coordination of two hemispheres, supporting that
1) higher cognitive function is related to functional segregation proposed in Chapter 4 and 2) the
important role of information transfer through interhemispheric connectivity toward cognition
stated in Chapter 3.
176
6.4.4 Disease severity and comorbidities are related to compensated dynamics in PD
The other pattern in our results demonstrated that higher disease severity and comorbidities
(depression and apathy) were associated with higher global dynamics and connectivity variation
in intra-hemispheric connections as well as better language functions measured with sentence
repetition and verbal fluency, which possibly indicated that increased dFC compensated for
functional disruptions elsewhere. These increased dynamics may therefore facilitate language
processes, causing paradoxical associations between higher disease severity and better relative
language functions.
Although the exact mechanisms of altered language function in PD remains unclear [McNamara
and Durso, 2018], novel analysis investigating dFC has revealed that flexible and dynamic
networks facilitate language function [Chai et al., 2016], which explained the inter-correlations
between language performance and increased dynamic features especially dFC in intra-
hemispheric connections in our results. In addition, the language sub-test in MoCA actually
reflects both language ability and executive function because verbal fluency is one of the tasks. As
previously discussed, dFC has been associated with executive function in several studies [Mattar
et al., 2015; Nomi et al., 2017], the findings further strengthened the conclusion that better
language/executive function in PD is associated with increased dFC features rather than disease
severity and comorbidities directly.
Interestingly, among the 4 data sets in the MCCA, clinical and stationary FC sets expressed the
strongest correlation compared to other sets (r=0.8). UPDRS scores, depression and apathy
symptoms were strongly anti-correlated with functional segregation measured with transitivity.
The loadings of these two sets supported the conclusion in the previous chapters that disease
177
severity and comorbidity are related to functional segregation, and FC in PD has shifted toward a
segregation-orientated brain organization (please see Chapter 4).
6.4.5 Limitations and future work
There are several limitations to our study. All MS subjects were relapsing remitting in the study.
Therefore, the results might not be representative enough to the whole MS population, but the
connectivity profile related to cognition in the RRMS disease stage was suggested. Several MS
subjects were considered outliers and removed from the analysis, which reduced the sample size.
In PD, as the sample size was relatively small for machine-learning approaches, we only included
the variables that we were mostly interested in in order to avoid overfitting the data. Therefore,
some effects of demographic and clinical features were not assessed. A previous chapter discussed
heavily on gender effects to cognition in PD and MS, but we were unable to evaluate such an effect
in this study. During the analysis, if gender was included in the demographics data set, none of the
components were significant and the correlation coefficients between data sets were small
(r=0.01~0.4). Perhaps, compared to gender, cognitive function was more associated with other
variables such as education, disease severity, and fMRI measures in a multivariate relation.
As the original study design focused on the cognitive impairments commonly seen in the clinic,
the selected ROIs covered the areas that have been identified in the traditional neuropsychological
literature. However, with a greater recognition that complex cognition is a network phenomenon,
complete whole brain coverage should be included in the future research. The issue of window
length and the ways to report connectivity changes might be confounding factors while
investigating dynamic functional connectivity. In this study, we used suggested window length
parameters and selected six network features to summarize connectivity differences, which might
178
not capture all the potential connectivity changes. Recently other approaches have been proposed
to study network dynamics in an unbiased manner, such as the Sticky Weighted Regression Model
[Liu et al., 2015] and Dynamic Conditional Correlations [Lindquist et al., 2014]. Furthermore,
other methods have been applied to report meaningful connectivity patterns other than selecting
certain features [Leonardi et al., 2013]. These methods should also be tested with clinical data in
the future. We included graphical measures as stationary features, but other network
representations such as whole-brain connectivity matrices are also alternatives. In addition, nine
MS subjects were considered outliers, implying inter-subject variability may still be an issue for
robustness. Finally, changes in functional connectivity might be a result of compensation and
adaptation for structural damages in MS. We did include brain volume and lesion load to test
whether structural damages had impact on the results, but the MCCA model was not significant
(data are not shown). Although structural damage is not commonly considered in PD, recent
studies have discovered white matter microstructural changes with novel analyses [Zeighami et
al., 2017]. Combining structural MRI data and functional connectivity in a sophisticated way is
beyond the scope of this thesis, but it will be implemented to further investigate the disease effects
structurally and functionally.
6.5 Conclusion
In this study, we performed a sliding window approach and a graph theoretical analysis on rsfMRI
data to investigate dynamic functional connectivity and stationary network characteristics. MCCA
was applied to explore the associations between dynamic, static network measures and behavioural
data which included demographics, clinical data, cognitive performances, and affective status in
subjects with RRMS and PD.
179
With an MCCA approach that controlled for age effects, we discovered that better executive
functioning in MS was supported by higher education, stronger rsFC, and dynamic functional
connectivity; whilst disease severity was highly related to poor executive functioning and affective
variables, reinforcing the strong relationship between pathology and comorbidities. In PD, the
inter-correlations in MCCA revealed that better memory function was related to segregated brain
networks and dynamics of interhemispheric connections, reinforcing the role of interhemispheric
connectivity to cognition; while disease severity and comorbidity were associated with increased
dFC in global and intra-hemispheric connections, suggesting a compensatory effect and perhaps
causing the paradoxical association between disease severity and language performance.
The brain-behaviour relations revealed in this study may provide influential information for the
development of customized cognitive treatment and rehabilitation in neurological conditions (e.g.
rehab in specific domains based on disease severity), but further research is necessary to better
understand the disease effects as a whole such as combining structural, functional, and behavioural
data.
180
Chapter 7: Conclusion
The research presented in this dissertation utilized novel approaches to study resting-state
functional connectivity (rsFC) and the associations to cognitive deficits in two distinct
neurological conditions: Parkinson’s Disease (PD) and Multiple Sclerosis (MS). The goals were
to investigate whether advanced network measures have potential to serve as biomarkers, explore
relations between altered rsFC and cognitive performance, and predict cognitive function and
disease severity based on rsFC. The knowledge provides important insights to the understanding
of disease effects and treatment development.
7.1 Summary of the research
The included data in each chapter are summarized in Table 7.1.
Chapter Subjects Age Cognitive tests Project
Ch2 31 PD
46 MS
PD: 59.9±10.8(f), 61.5±9.4(m)
MS: 41.1±10.3(f), 45.8±11.6(m) Comprehensive
battery
PD: PPMI
MS: COGMS
Ch3 12 PD & 10 NC
25 MS & 41 NC
PD: 60.0±9.9, NC: 59.4±6.1
MS: 37.2±9.5, NC: 34.9±10.1
PD: limited
MS:
PASAT/SDMT
PD: BCT
MS: OPERA
Ch4 23 PD & 19 NC
(8 showed artifacts)
18 MS & 15 NC
46 MS for BBA
PD: 60.95±9.7, NC: 56.13±16.9
MS: 32.00±4.9, NC: 28.93±5.0
46 MS: 42.89±10.9
Comprehensive
battery as ch2
PD: PPMI
MS: COGMS
Ch5 24 PD & 15 NC
18 MS & 15 NC
46 MS for BBA
PD: 68.38±4.7, NC: 69±4.8
MS: same as ch4
PD: MoCA sub-
scores
MS: as ch2
PD: GFM2
MS: COGMS
Ch6 24 PD
37 MS
(9 outliers)
same as ch5
PD: as ch5
MS: as ch2,4,5
PD: GFM2
MS: COGMS
Table 7.1 The summary of data used in each chapter.
181
[BBA: brain-behaviour analysis]
7.1.1 Cognitive impairments and functional connectivity
Chapter 1 presented a literature review on cognitive deficits in PD and MS – two distinct
neurological diseases that both demonstrate similar cognitive deficits especially executive
dysfunction – as well as a summary of the clinical research which investigates rsFC and cognitive
impairments with both traditional (i.e. seed-based connectivity and independent component
analysis-derived resting networks) and novel/advanced approaches (i.e. graph theory analysis and
time-varying approaches). This chapter concluded that there are common causes and unique
features for executive impairments in PD and MS. Impaired long-range connections potentially
impact the “input component” of executive function, which is a common feature in both diseases.
Decreased efficiency of global neuronal transfer has been commonly reported in MS, which might
be a result of hub failure. In PD, the frontal-striatal loops affected by dopamine depletion affects
the “output component” of the executive function and hub reorganization (i.e. non hubs become
more important and hubs lose the central role in the networks) altered brain organization. In
addition, as cognitive inflexibility has been commonly seen in PD clinically, the chapter also
hypothesizes that connectivity flexibility might be impaired (i.e. dynamic functional connectivity
which facilitates higher-order cognition) as well and connectivity flexibility shall share certain
degrees of association with cognitive flexibility.
7.1.2 Cognitive profiles and the associations to demographical features
In Chapter 2, Canonical Correlation Analysis (CCA), a multivariate approach, revealed complex
inter-correlated patterns between cognitive performance and demographical characteristics to
compare PD and MS in terms of cognitive deficits reported in the literature. In MS, female gender
182
was associated with better executive functions such as set-shifting ability; while male gender and
disease severity were related to poor performance in card sorting tests. In PD, female gender and
education were associated with better performance in almost all the cognitive domains that are
commonly impaired such as verbal learning and memory, executive function, attention, and
processing speed. On the other hand, age and depression were anti-correlated with better cognitive
abilities.
Taken together, the chapter reveals an inter-correlated pattern between cognitive performance,
clinical evaluation, and demographical characteristics in PD and MS. Moreover, this study
concluded that gender is an influential factor in preserving cognitive function regardless of
pathology, possibly due to the purported neuroprotective effects of estrogen.
7.1.3 Long-range connections, interhemispheric connectivity, and cognitive function
As Chapter 1 hypothesized that long-range connections might be the common cause to cognitive
deficits, Chapter 3 utilized whole brain connectivity analysis to investigate long-range connection
as well as interhemispheric functional connectivity. Several machine learning methods were
carried out to explore the relations between rsFC and cognitive performance.
In MS, interhemispheric connectivity appeared to be enhanced and homogenized, possibly
indicating a compensatory effect. With robust and LASSO regression, we reported that adequate
cognitive performance requires distributed homologous brain regions, and that interhemispheric
connectivity can accurately predict performance of attention, processing speed, and working
memory functions. With an exploratory approach, frontal-parietal/occipital connections were
significantly different between controls and MS and such pattern was correlated with cognitive
183
performance that required attention and memory, supporting the important role of long-range
connections in cognitive processes.
In PD, enhanced interhemispheric connectivity was observed in the medial prefrontal cortex,
possibly as a compensatory mechanism. Only dopaminergic medication dose could be mildly
predicted by the altered interhemispheric connectivity. In addition to interhemispheric
connectivity, several connections within frontal-striatal loops showed differences in PD. Although
they were part of the “output component” of executive function, no significant associations with
behavioural scores were found, possibly due to the fact that the patients were very mild and none
showed significant cognitive impairment.
Overall, long-range connections and interhemispheric connectivity could predict cognitive
performance in MS, which supported the hypothesis in Chapter 1. However, the brain-behaviour
associations in PD were not very promising. Perhaps, other aspects of rsFC should be taken into
account such as dynamic functional connectivity.
7.1.4 Topological brain organization and executive function
In Chapter 4, graph theoretical approaches were applied to capture rsFC network characteristics.
As functional integration and segregation are important principles of brain organization, we mainly
calculated the measures representing these two characteristics. In addition, as hubs have been
associated with higher-order cognitive function, measures of hub structures were included as well.
In order to explore the brain-behaviour relations reflecting how brain organization is associated
with cognitive function, we utilized several methods such as correlation, regression, and machine
learning approaches.
184
In PD, although global measures did not show any differences, several local measures
demonstrated alterations across cortices, indicating 1) a segregation-oriented brain organization
and 2) hub vulnerability. Moreover, increased functional segregation was correlated with better
cognitive performance which requires executive skills as well as verbal learning and memory
abilities. The observed brain-behaviour relations further supported that FC has become more
segregated in order to support cognitive function in PD.
In MS, there were no alterations in brain organization, which might due to the mild disease course
in our cohort. However, interestingly, brain-behaviour analyses revealed promising results. First,
modularity, a measure of functional segregation, was related to the performance of verbal fluency
test, indicating that executive skills were associated with functional segregation in MS. In addition,
in several linear regression models, disease severity, psychiatric comorbidities, and cognitive
performance in executive and processing speed domains could be significantly predicted by local
measures located across cortices and the cerebellum. Moreover, with a machine learning approach,
such predictability was enhanced between local efficiency and disease duration. Taken together,
the brain-behaviour associations illustrated that the disease was affected/predicted by distributed
regions and executive skills were significantly associated with the coordination between not only
frontal, temporal, and parietal regions but also cerebellum, supporting the “cognitive role” of
cerebellum in neurological diseases.
In short, brain organization reflected information transfer in a topological fashion, which allowed
us to investigate how diseases altered communication between brain regions. In PD, we
demonstrated hub vulnerability, which is consistent with previous studies not only in PD but other
185
neurological and psychiatric disorders. In MS, we revealed robust brain-behaviour relations and
highlighted the influence of distributed brain regions to both disease course and cognitive function.
7.1.5 Dynamic functional connectivity and executive function
In Chapter 5, a sliding window approach with inverse covariance matrix estimation was used to
estimate dynamic functional connectivity with features representing dynamics in global and
specific connections. Correlation was carried out to explore the associations between dynamics
features and behavoural data (demographics, clinical measures, and cognitive scores). Linear
regression was applied to study the relations between principal components of dynamic features
and behavioural data.
In MS, global dynamics and connectivity strength/density were decreased, but interhemispheric
connections increased their dynamics, suggesting a compensatory mechanism whereby
interhemispheric connections become more variable to compensate for global functional
disruption. Longer disease duration was related to decreased dFC in several specific connections
and could be predicted by principal components of dynamic features. Although components
representing main effects of interhemispheric dynamics could predict set-shifting ability in a
regression model, this relation was not representative of the whole data set.
In PD, none of the dynamics features were different between healthy subjects and patients, which
could be explained by the fact that all PD subjects were very mildly affected. However, subtle
changes of dFC were related to cognitive performance. Higher dynamics in interhemispheric
connections were related to better performance in a memory test, emphasizing the importance of
dynamic coordination between two hemispheres in cognitive processes. Two regression models
186
showed significant brain-behaviour associations. A principal component that represented overall
dynamics could predict disease severity; while age and a component which contrasted effects from
interhemispheric vs non-interhemispheric dFC could predict memory function. The regression
results supported the interpretation of the correlation patterns but with stronger robustness.
To conclude, with dynamic features derived from the analysis, we observed compensatory effects
in MS, whereby increased dynamics in interhemispheric connections compensated for reduced
dynamics in global networks. Furthermore, dynamic connectivity patterns could predict disease
duration. In PD, although no network changes were observed, subtle dFC alterations were related
to cognitive performance, highlighting the role of dFC to cognition and supporting the idea that
dFC may be a key role in PD, as proposed in Chapter 1.
7.1.6 Combining dynamic, stationary functional connectivity and behavioural data
reveals complementary information
As mentioned in Chapter 1, more sophisticated statistical approaches are needed to explore brain-
behaviour associations. In chapter 6, Multiset Canonical Correlation Analysis (MCCA), a
multivariate and machine learning approach, was applied to investigate the relations between
different aspects of FC (i.e. stationary and dynamic), demographics, clinical data, and cognitive
scores jointly. The dynamic FC set contained dynamic features derived from a sliding window
approach as described in Chapter 5. For the stationary set, graph theory measures were calculated
to represent functional integration, segregation, and hub structures as mentioned in Chapter 4. This
multivariate approach allowed us to combine different modalities to come up with complementary
information to understand disease effects clinically, functionally, and cognitively.
187
In MS, five data sets were included in the MCCA: dynamic and stationary rsFC, demographics,
cognitive scores, and affective variables. Better cognitive performance in several tests were
correlated with higher dynamics, stronger static connectivity strength, and longer education; whilst
worse set-shifting abilities were related to denser static connectivity, higher disease severity,
longer disease duration, and comorbidities such as depression, anxiety, and fatigue. Based on these
inter-correlated patterns, we suggest that the balance between dynamic and stationary connectivity
and coordination of information flow might be the key factor to facilitate cognition. In addition,
this MCCA pattern also indicated a co-existence of disease severity, psychiatric comorbidity,
reduced shifting abilities, mental speed, and denser but rigid static rsFC in MS.
In PD, the MCCA model utilized four data sets: dynamic and stationary rsFC, clinical data, and
cognitive performance measured in MoCA sub-tests. Better memory function was associated with
higher dynamics in interhemispheric, graphical measures assortativity and transitivity, supporting
the conclusion in the previous chapters that 1) dynamic coordination between two hemispheres
was important for cognitive function, and 2) rsFC in PD shifts toward a more segregation-oriented
brain organization. On the other hand, better language performance was related to higher dynamics
in global and intrahemispheric FC, higher disease severity, more severe depression and apathy
symptoms. In this brain-behaviour pattern, perhaps higher dynamics represented compensatory
effects and therefore facilitated language function. As disease severity, depression, and apathy
jointly are less likely to improve cognitive function, we propose that better language performance
might be due to compensated higher dynamics.
Using MCCA with different modalities to examine the brain-behaviour relations provided
complementary information for the development of customized cognitive treatment and
188
rehabilitation in neurological conditions. Although choosing the data sets included in MCCA could
be challenging, this approach offers another opportunity to assess brain-behaviour relations more
comprehensively and supports the benefits of using advanced methods mentioned in Chapter 1.
7.2 Strengths and limitations
7.2.1 Strengths: functional connectivity analysis
Unlike the traditional clinical research in PD and MS, this thesis research carried out whole brain
connectivity analysis with a focus on brain organization and dynamic functional connectivity as
opposed to targeting front-striatal loops in PD and independent component analysis (ICA) derived
resting-state networks in MS. With graphical analysis, we have a border view on how brain
organization is altered and how connections between regions are changed as a whole. Moreover,
with whole brain analyses, there is a better chance to investigate compensatory mechanisms as it
is unclear whether and where these effects appear. Additionally, with a time- varying approach,
we were able to study how connectivity changes temporally and whether such temporal transitions
were related to specific aspects of the disease. Traditional analysis methods assume that
connectivity remains relatively constant, neglecting the variable and fluctuating nature of the brain.
Although how to report connectivity variation remains an issue with the sliding window approach,
the current method in the thesis research shed light on dynamic nature of information transfer in
the brain. Taken together, the approaches utilized in this thesis research were able to study the
system-level neurophysical events in the brain.
189
7.2.2 Strengths: tested domains
Several tests were used throughout the studies to evaluate patients’ cognitive function and the
tested domains were the commonly affected domains in PD and MS. Although these tested
domains were not exactly the same between two populations, executive function was a common
theme in both PD and MS. Throughout the thesis research, several domains have been associated
with demographics and fMRI measures. In chapter 2, all the tested domains (visuospatial,
executive function, verbal learning and memory, processing speed, working memory) were related
to demographics in PD; while executive function was most significantly affected domain in MS.
In chapter 3, due to the original study design, no tested domains were significant in PD; while
processing speed and working memory were associated with fMRI measures in MS. In chapter 4,
processing speed was associated with fMRI graphical measures in PD; while executive function is
the only significant variable in MS. In chapter 5, due to study design (i.e. only MoCA sub-scores
were used to assess cognition), only memory was strongly related to dynamic fMRI measures
(relevant discussion in section 5.4.4); while no cognitive domains were significantly related to
dynamic features in MS. In chapter 6, memory, language, and executive function were correlated
with FC in PD; while working memory, processing speed, and executive function were associated
with FC in MS. Overall, executive function has been associated with demographics and fMRI
measures in three of the chapters. In the remaining two chapters where overall executive function
was not reported, working memory, processing speed, and memory retrieval function were
specifically assessed. These cognitive functions all required certain degree of executive skills to
achieve the oriented behaviour and are routinely considered to subserve higher-order cognitive
function [Miller and Wallis, 2009]. Therefore, the results still reflected how brain networks and
190
connections facilitated executive skills in PD and MS. To conclude, with whole brain connectivity
analysis and the tested domains reported in the thesis, the goal to probe the systems-level
neurobiological bases for executive function in patient populations has been achieved.
7.2.3 Strengths: brain-behaviour analysis
Many studies use cognitive scores to evaluate the cognitive states of patients rather than looking
into how task performance is related to functional connectivity. A few studies have implemented
statistical approaches such as correlation analysis to explore the associations between connectivity
and cognitive function; however, as we have proposed, univariate methods may not be able to
reveal complex interactions and results are either unpromising or hard to interpret. In this
dissertation, several chapters reported brain-behaviour associations with both univariate (e.g.
correlation, regression) and multivariate approaches (e.g. PCA, MCCA, CCA). The multivariate
approaches take into account the complex inter-correlated nature of human brain and behaviour,
which is especially important for diseased populations due to disease heterogeneity. Therefore, the
use of multivariate/machine learning methods in this thesis research is significant.
7.2.4 Limitations: study design
There are several limitations to the studies. First, multivariate and machine learning approaches
usually require a big sample size. However, with neuroimaging data, it is more costly and time
consuming to acquire a huge sample compared to other types of data. Therefore, very often one
may over-fit the data. Although some common strategies potentially solve this problem (such as
perform the analysis in a leave-one-out fashion, reduce data dimension first, etc.), overfitting data
is still a big problem given the sample size in our cohort. In any future study, a power analysis
191
would be beneficial to estimate the required sample size, now that the current studies could provide
an approximation for effective size. Sample size of training data is one of the key factors in
generating efficient predictors/classifiers, especially for some of the machine learning approaches
utilized here [Varoquaux and Thirion, 2014]. In our preprocessing steps, we used standard
approaches and did not perform extra steps to remove physiological noise caused by heart beat and
respiratory rhythm, which may or may not affect connectivity analysis. Therefore, in the
connectivity analysis, partial correlation or regularized partial correlation were preferred as they
tend to be more robust to noise compared to pairwise correlation. Furthermore, as highlighted in
the thesis title, one of the main focuses is executive function even though executive function and
cognitive function are both used throughout the chapters. In fact, it is difficult to clearly separate
executive function from other cognitive abilities. Many cognitive processes require more or less
executive skills and the neuropsychological tests in the research may not reflect pure executive
functioning. Yet the main focus still lay on higher-order cognitive processes which require
intensive thinking and planning in opposed to lower-order functions such as vision, long-term
memory, and motor as discussed in section 7.2.2.
7.2.5 Limitations: gender issues
In chapter 2, we found that gender appears to be a strong influential factor for cognition in both
disease populations; however, this factor was less prominent in the rest of the chapters. Gender
was mostly included in multivariate analyses such as CCA in chapter 2 and MCCA in chapter 6.
As previously mentioned in section 6.4.5, including gender in MCCA did not provide significant
or robust results. Based on results in chapter 2 and 6, gender is indeed an influential factor in
cognition when no imaging modalities are included; however, when fMRI measures are included,
192
gender effects are overwhelmed by FC measures. Taken together, we conclude that cognition is
likely affected by gender, but functional connectivity, which underlines neural activity in a
macroscale, is a more powerful factor to cognition compared to gender. This may because
functional connectivity itself is strongly affected by gender [Gong et al., 2011].
7.2.6 Limitations: compensatory effects
In disease populations, increased FC or expanded activation map is frequently interpreted as
compensation. However, whether the observed alterations represent successful compensation or
(mal)adaptation remains debated [Scheller et al., 2014]. Simply demonstrating changes of fMRI
measures without supporting by concomitant behavioural changes may not necessarily represent
successful compensation. Instead, this could be a process of maladaptation (i.e. disinhibition)
[Schoonheim et al., 2015]. Throughout the research chapters, we also discovered connectivity
changes that were correlated with better performance, indicating that these changes likely served
as compensatory mechanisms. For example, in chapter 3, increased interhemispheric connectivity
was related to better performance of tasks probing processing speed and working memory in MS.
In chapter 4, increased functional segregation was also associated with better performance of
processing speed in PD; while subtle changes of segregation was related to executive function in
MS. Hub reorganization in PD may indicate compensation, whereby hub regions lost their central
role in the network and non-hub regions, which may still receive dopamine from the ventral
tegmental area, became more important. In chapter 5, interhemispheric connections became more
dynamic to possibly compensate for decreased global measures in MS. Although the brain-
behaviour association did not explain the majority of the variance in the data, such increased
interhemispheric dynamics was related to better executive function in MS (Appendix D). Overall,
193
our observations of altered connectivity were supported by brain-behaviour relations, allowing us
to conclude that these connectivity changes were related to compensation. Since the included
cohorts in the studies were all mildly affected and most of the subjects did not demonstrate severe
cognitive impairments, the compensatory effects in the thesis only represent adaptation at early
stage before network collapse [Schoonheim et al., 2015]. In order to comprehensively study
compensation, more imaging data, bigger sample size, and subjects at different disease stage would
be required.
7.3 Future research directions
One of the biggest goals in the research is to define the connections, networks, and/or brain regions
that are important to cognitive dysfunction in neurological disorders. With the knowledge, it would
be beneficial to develop efficient treatment for cognitive deficits. For example, as Chapter 3 and 4
revealed, long-range connections, interhemispheric connections, and hub regions have potential to
predict cognitive functions. A cognitive rehabilitation strategy that specifically targets these
regions and networks might provide a more efficient treatment. As the current pharmaceutical
approaches to treat cognitive dysfunction in PD and MS produce only modest improvements,
cognitive training as a rehabilitation strategy could be an alternative. In fact, such strategy has been
applied to MS subjects. Although there are some promising results, the overall literature is about
such interventions [Mitolo et al., 2015]. However, implementing the knowledge of network effects
to cognition may provide insights to design more effective interventions.
For future research, there are several possible directions. First, different models of connectivity
analysis are desirable. The current research utilized the most common ones in neuroscience field,
but models with different technical and mathematical assumptions might be able to reveal hidden
194
patterns in connectivity. Moreover, in this thesis research, we used resting-state data and tried to
link rsFC with cognitive performance outside of the scanner as patients are more able to cope with
resting-state scans than tasks. Although it has been shown that rsFC reflects neuronal activity of
cognitive tasks [Smith et al., 2009], task-driven fMRI may be more suitable to investigate brain-
behaviour associations of executive function. Furthermore, there has been a tremendous interest
in linking functional and structural connectivity. In MS, for example, it would be interesting to
know whether functional connectivity is related to lesions and/or other structural changes. In PD,
it would be challenging but exciting to explore whether microstructural changes affect dynamic
functional connectivity. With both functional and structural information, together with clinical
data and cognitive scores, scientists will be able to understand diseases structurally, functionally,
clinically, and cognitively.
195
Bibliography
Aarsland D, Creese B, Politis M, Chaudhuri KR, Ffytche DH, Weintraub D, Ballard C (2017):
Cognitive decline in Parkinson disease. Nat Rev Neurol 13:217–231.
Achard S, Salvador R, Whitcher B, Suckling J, Bullmore E (2006): A Resilient, Low-Frequency,
Small-World Human Brain Functional Network with Highly Connected Association
Cortical Hubs. J Neurosci 26:63–72.
Achiron A, Chapman J, Tal S, Bercovich E, Gil H, Achiron A (2013): Superior temporal gyrus
thickness correlates with cognitive performance in multiple sclerosis. Brain Struct Funct
218:943–50.
Alamri Y, Pascoe M, Dalrymple-Alford J, Anderson T, MacAskill M (2017): Through new eyes:
gaze analysis of Parkinson disease patients performing the Symbol Digit Modalities Test.
Neurology 88:[Epub Ahead].
Allen E a., Damaraju E, Plis SM, Erhardt EB, Eichele T, Calhoun VD (2014): Tracking whole-
brain connectivity dynamics in the resting state. Cereb Cortex 24:663–676.
Alonso Recio L, Martín P, Carvajal F, Ruiz M, Serrano JM (2013): A holistic analysis of
relationships between executive function and memory in Parkinson’s disease. J Clin Exp
Neuropsychol 35:147–159.
Alosaimi FD, AlMulhem A, Moscovici M, AlShalan H, Alqazlan M, Aldaif A, Sockalingam S
(2017): The Relationship between Psychosocial Factors and Cognition in Multiple
Sclerosis. Behav Neurol 2017.
Alvarez JA, Emory E (2006): Executive function and the frontal lobes: A meta-analytic review.
Neuropsychol Rev 16:17–42.
Amboni M, Tessitore A, Esposito F, Santangelo G, Picillo M, Vitale C, Giordano A, Erro R, de
196
Micco R, Corbo D, Tedeschi G, Barone P (2014a): Resting-state functional connectivity
associated with mild cognitive impairment in Parkinson’s disease. J Neurol 262:425–434.
Amboni M, Tessitore A, Esposito F, Santangelo G, Picillo M, Vitale C, Giordano A, Erro R, de
Micco R, Corbo D, Tedeschi G, Barone P (2014b): Resting-state functional connectivity
associated with mild cognitive impairment in Parkinson’s disease. J Neurol 262:425–434.
Ansari D (2012): Culture and education: New frontiers in brain plasticity. Trends Cogn Sci
16:93–95.
Augustine EF, Pérez A, Dhall R, Umeh CC, Videnovic A, Cambi F, Wills AMA, Elm JJ, Zweig
RM, Shulman LM, Nance MA, Bainbridge J, Suchowersky O (2015): Sex differences in
clinical features of early, treated Parkinson’s disease. PLoS One 10:1–11.
Baggio H-C, Sala-Llonch R, Segura B, Marti M-J, Valldeoriola F, Compta Y, Tolosa E, Junqué
C (2014): Functional brain networks and cognitive deficits in Parkinson’s disease. Hum
Brain Mapp 35:4620–34.
Baggio H, Segura B, Junque C, de Recus M, Sala-Llonch R, Van den Heuvel M (2015): Rich
Club Organization and Cognitive Performance in Healthy Older Participants. J Cogn
Neurosci 27:1801–1810.
Bailey CE (2007): Cognitive accuracy and intelligent executive function in the brain and in
business. Ann N Y Acad Sci 1118:122–141.
Ballinger EC, Ananth M, Talmage DA, Role LW (2016): Basal Forebrain Cholinergic Circuits
and Signaling in Cognition and Cognitive Decline. Neuron 91:1199–1218.
Barkhof F (2002): The clinico-radiological paradox in multiple sclerosis revisited. Curr Opin
Neurol 15:239–245.
Barone P, Santangelo G (2010): Interaction between affect and executive functions in
197
Parkinson’s disease. In: Emre, M, editor. Cognitive Impairment and Dementia in
Parkinson’s Disease. Oxford University Press. pp 66–73.
Baumeister TR, Lin S-J, Vavasourc I, Kolind S, Kosaka B, Li DKB, Traboulsee A, MacKay A,
McKeown MJ Multiset canonical correlation analysis reveals complex associations between
myelination, cognitive performance, and clinical characteristics in multiple sclerosis. Prep.
Beatty WW, Aupperle RL (2002): Sex differences in cognitive impairment in multiple sclerosis.
Clin Neuropsychol 16:472–480.
Benedict R, Bruce J, Dwyer M, Abdelrahman N, Hussein S, Weinstock-Guttman B, Garg N,
Munschauer F, Zivadinov R (2006): Neocortical atrophy, third ventricular width, and
cognitive dysfunction in multiple sclerosis. Arch Neurol 63:1301–1306.
Benedict RHB, Zivadinov R (2011): Risk factors for and management of cognitive dysfunction
in multiple sclerosis. Nat Rev Neurol 7:332–342.
Bergendal G, Martola J, Stawiarz L, Kristoffersen-Wiberg M, Fredrikson S, Almkvist O (2013):
Callosal atrophy in multiple sclerosis is related to cognitive speed. Acta Neurol Scand
127:281–289.
Berman BD, Smucny J, Wylie KP, Shelton E, Kronberg E, Leehey M, Tregellas JR (2016):
Levodopa modulates small-world architecture of functional brain networks in Parkinson’s
disease. Mov Disord 31:1676–1684.
Bertolero MA, Yeo BTT, D’Esposito M (2015): The modular and integrative functional
architecture of the human brain. Proc Natl Acad Sci U S A 112:E6798-807.
Betzel RF, Fukushima M, He Y, Zuo XN, Sporns O (2016): Dynamic fluctuations coincide with
periods of high and low modularity in resting-state functional brain networks. Neuroimage
127:287–297.
198
Blair C, Peters Razza R (2007): Relating Effortful Control, Executive Function, and False Belief
Understanding to Emerging Math and Literacy Ability in Kindergarten 78:647–663.
Bloom J, Hynd G (2005): The role of the corpus callosum in interhemispheric transfer of
information: excitation or inhibition? Neuropsychol Rev 15:59–71.
Bodini B, Cercignani M, Khaleeli Z, Miller D, Ron M, Penny S, Thompson A, Ciccarelli O
(2013): Corpus callosum damage predicts disability progression and cognitive dysfunction
in primary-progressive MS after five years. Hum Brain Mapp 34:1163–1172.
Bohnen NI, Kaufer DI, Hendrickson R, Ivanco LS, Lopresti BJ, Constantine GM, Mathis CA,
Davis JG, Moore RY, DeKosky ST (2006): Cognitive correlates of cortical cholinergic
denervation in Parkinson’s disease and parkinsonian dementia. J Neurol 253:242–247.
Bonavita S, Gallo A, Sacco R, Corte M Della, Bisecco A, Docimo R, Lavorgna L, Corbo D,
Costanzo A Di, Tortora F, Cirillo M, Esposito F, Tedeschi G (2011): Distributed changes in
default-mode resting-state connectivity in multiple sclerosis. Mult Scler 17:411–422.
Bonelli RM, Cummings JL (2007): Frontal-subcortical circuitry and behavior. Dialogues Clin
Neurosci 9:141–151.
Bowman F (2014a): Brain Imaging Analysis. Annu Rev Stat Appl 1:61–85.
Bowman FD (2014b): Brain Imaging Analysis. Annu Rev Stat its Appl 1:61–85.
Brann DW, Dhandapani K, Wakade C, Mahesh VB, Khan MM (2007): Neurotrophic and
neuroprotective actions of estrogen: Basic mechanisms and clinical implications. Steroids
72:381–405.
Braun U, Plichta MM, Esslinger C, Sauer C, Haddad L, Grimm O, Mier D, Mohnke S, Heinz A,
Erk S, Walter H, Seiferth N, Kirsch P, Meyer-Lindenberg A (2012): Test-retest reliability of
resting-state connectivity network characteristics using fMRI and graph theoretical
199
measures. Neuroimage 59:1404–1412.
Braun U, Schäfer A, Walter H, Erk S, Romanczuk-Seiferth N, Haddad L, Schweiger JI, Grimm
O, Heinz A, Tost H, Meyer-Lindenberg A, Bassett DS (2015): Dynamic reconfiguration of
frontal brain networks during executive cognition in humans. Proc Natl Acad Sci U S A
112:11678–83.
Bressler S, Scott Kelso JA (2001): Cortical coordination dynamics and cognition. Trends Cogn
Sci 5:26–36.
Brown TE, Landgraf JM (2010): Improvements in executive function correlate with enhanced
performance and functioning and health-related quality of life: evidence from 2 large,
double-blind, randomized, placebo-controlled trials in ADHD. Postgrad Med 122:42–51.
Bruce JM, Arnett PA (2005): MS patients with depressive symptoms exhibit affective memory
biases when verbal encoding strategies are suppressed. J Int Neuropsychol Soc 11:514–521.
Buckner RL (2013): The cerebellum and cognitive function: 25 years of insight from anatomy
and neuroimaging. Neuron 80:807–815.
Bullmore E, Sporns O (2012): The economy of brain network organization. Nat Rev Neurosci
13:336–349.
Bunge S, Souza M (2009): Executive Function and Higher-Order Cognition : Neuroimaging.
Encycl Neurosci 4:111–116.
Carpenter PA, Just MA, Reichle ED (2000): Working memory and executive function: evidence
from neuroimaging. Curr Opin Neurobiol 10:195–199.
Chai LR, Mattar MG, Blank IA, Fedorenko E, Bassett DS (2016): Functional Network Dynamics
of the Language System. Cereb Cortex 26:4148–4159.
Chang C, Glover GH (2010): Time-frequency dynamics of resting-state brain connectivity
200
measured with fMRI. Neuroimage 50:81–98.
Chang C, Liu Z, Chen MC, Liu X, Duyn JH (2013): EEG correlates of time-varying BOLD
functional connectivity. Neuroimage 72:227–236.
Chen T, Cai W, Ryali S, Supekar K, Menon V (2016a): Distinct Global Brain Dynamics and
Spatiotemporal Organization of the Salience Network. PLoS Biol 14:1–21.
Chen X, Wang ZJ, McKeown M (2016b): Joint Blind Source Separation for Neurophysiological
Data Analysis: Multiset and multimodal methods. IEEE Signal Process Mag 33:86–107.
Chiang S, Cassese A, Guindani M, Vannucci M, Yeh HJ, Haneef Z, Stern JM (2016): Time-
dependence of graph theory metrics in functional connectivity analysis. Neuroimage
125:601–615.
Chiaravalloti ND, DeLuca J (2008): Cognitive impairment in multiple sclerosis. Lancet Neurol
7:1139–1151.
Chiaravalloti ND, Ibarretxe-Bilbao N, Deluca J, Rusu O, Pena J, García-Gorostiaga I, Ojeda N
(2014): The source of the memory impairment in Parkinson’s disease: Acquisition versus
retrieval. Mov Disord 29:765–771.
Chung HJ, Weyandt LL, Swentosky A (2014): The Physiology of Executive Functioning. In:
Goldstein, S, Naglieri, JA, editors. Handbook of Executive Functioning. Springer New
York. pp 13–27.
Cohen JR, D’Esposito M (2016): The Segregation and Integration of Distinct Brain Networks
and Their Relationship to Cognition. J Neurosci 36:12083–12094.
Cools R, Barker RA, Sahakian BJ, Robbins TW (2001): Mechanisms of cognitive set flexibility
in Parkinson’s disease. Brain 124:2503–2512.
Corkin S, Growdon JH, Locascio JJ (2003): Relation Between Clinical Characteristics of
201
Parkinson’s Disease and Cognitive Decline. J Clin Exp Neuropsychol 25:94–109.
Correa NM, Li YO, Adali T, Calhoun VD (2008): Canonical correlation analysis for feature-
based fusion of biomedical imaging modalities and its application to detection of associative
networks in Schizophrenia. IEEE J Sel Top Signal Process 2:998–1007.
Costa A, Monaco M, Zabberoni S, Peppe A, Perri R, Fadda L, Iannarelli F, Caltagirone C,
Carlesimo GA (2014): Free and cued recall memory in Parkinson’s disease associated with
amnestic mild cognitive impairment. PLoS One 9.
Cover K, Vrenkenb H, Geurtsb J, van Oosten BW, Jellesc B, Polmanc C, Stam C, van Dijk BW
(2006): Multiple sclerosis patients show a highly significant decrease in alpha band
interhemispheric synchronization measured using MEG. Neuroimage 29:783–788.
Crossley NA, Mechelli A, Scott J, Carletti F, Fox PT, Mcguire P, Bullmore ET (2014): The hubs
of the human connectome are generally implicated in the anatomy of brain disorders. Brain
137:2382–2395.
Cruz-Gómez ÁJ, Ventura-Campos N, Belenguer A, Ávila C, Forn C (2014): The link between
resting-state functional connectivity and cognition in MS patients. Mult Scler J 20:338–48.
D’Ambrosio A, Rocca MA, Valsasina P, Gallo A, De Stefano N, Pareto D, Barkhof F, Ciccarelli
O, Enzinger C, Tedeschi G, Stromillo ML, Arevalo M-J, Hulst H, Muhlert N, Koini M,
Filippi M (2018): Reduced Dynamism of Functional Connectivity Is Associated with
Cognitive Impairment in Multiple Sclerosis Patients: A Dynamic Functional Connectivity
Study in a Multi-Center Setting. Neurology 90:P3.379.
Damaraju E, Allen EA, Belger A, Ford JM, McEwen S, Mathalon DH, Mueller BA, Pearlson
GD, Potkin SG, Preda A, Turner JA, Vaidya JG, Van Erp TG, Calhoun VD (2014):
Dynamic functional connectivity analysis reveals transient states of dysconnectivity in
202
schizophrenia. NeuroImage Clin 5:298–308.
Damoiseaux JS, Rombouts SARB, Barkhof F, Scheltens P, Stam CJ, Smith SM, Beckmann CF
(2006): Consistent resting-state networks. Proc Natl Acad Sci U S A 103:13848–13853.
Dauer W, Przedborski S (2003): Parkinson ’ s Disease : Mechanisms and Models. Neuron
39:889–909.
Davis AS, Pierson EE, Finch WH (2011): A canonical correlation analysis of intelligence and
executive functioning. Appl Neuropsychol 18:61–68.
Deco G, Jirsa VK, McIntosh AR (2013): Resting brains never rest: Computational insights into
potential cognitive architectures. Trends Neurosci 36:268–274.
Deco G, Tononi G, Boly M, Kringelbach ML (2015): Rethinking segregation and integration:
contributions of whole-brain modelling. Nat Rev Neurosci 16:430–439.
Diamond A (2013): Executive functions. Annu Rev Psychol 64:135–168.
Díez-Cirarda M, Strafella AP, Kim J, Peña J, Ojeda N, Cabrera-Zubizarreta A, Ibarretxe-Bilbao
N (2018): Dynamic functional connectivity in Parkinson’s disease patients with mild
cognitive impairment and normal cognition. NeuroImage Clin 17:847–855.
Diez I, Erramuzpe A, Escudero I, Mateos B, Cabrera A, Marinazzo D, Sanz-Arigita EJ,
Stramaglia S, Cortes Diaz JM (2015): Information Flow Between Resting-State Networks.
Brain Connect 5:554–564.
Dirnberger G, Jahanshahi M (2013): Executive dysfunction in Parkinson’s disease: A review. J
Neuropsychol 7:193–224.
Disbrow EA, Carmichael O, He J, Lanni KE, Dressler EM, Zhang L, Malhado-Chang N,
Sigvardt KA (2014): Resting state functional connectivity is associated with cognitive
dysfunction in non-demented people with Parkinson’s disease. J Parkinsons Dis 4:453–465.
203
Dogonowski A, Andersen K, Madsen K, Sørensen P, Paulson O, Blinkenberg M, Siebner H
(2013): Multiple sclerosis impairs regional functional connectivity in the cerebellum.
Neuroimage Clin 4:130–138.
Donner R, Reiter M, Langs G, Peloschek P, Bischof H (2006): Fast active appearance model
search using canonical correlation analysis. IEEE Trans Pattern Anal Mach Intell 28:1690–
1694.
Douw L, Wakeman DG, Tanaka N, Liu H, Stufflebeam SM (2016): State-dependent variability
of dynamic functional connectivity between frontoparietal and default networks relates to
cognitive flexibility. Neuroscience 339:12–21.
Drake A, Weinstock-Guttman B, Morrow S, Hojnacki D, Munschauer F, Benedict R (2010):
Psychometrics and normative data for the Multiple Sclerosis Functional Composite:
replacing the PASAT with the Symbol Digit Modalities Test. Mult Scler J 16:228–237.
Duncan GW, Firbank MJ, Yarnall AJ, Khoo TK, Brooks DJ, Barker RA, Burn DJ, O’Brien JT
(2016): Gray and white matter imaging: A biomarker for cognitive impairment in early
Parkinson’s disease? Mov Disord 31:103–110.
Dzirasa K, Ramsey AJ, Takahashi DY, Stapleton J, Potes JM, Williams JK, Gainetdinov RR,
Sameshima K, Caron MG, Nicolelis MAL (2009): Hyperdopaminergia and NMDA receptor
hypofunction disrupt neural phase signaling. J Neurosci 29:8215–24.
Elliott R (2003): Executive functions and their disorders. Br Med Bull 65:49–59.
Eusebio A, Pogosyan A, Wang S, Averbeck B, Gaynor LD, Cantiniaux S, Witjas T, Limousin P,
Azulay JP, Brown P (2009): Resonance in subthalamo-cortical circuits in Parkinsons
disease. Brain 132:2139–2150.
Faivre A, Rico A, Zaaraoui W, Crespy L, Reuter F, Wybrecht D, Soulier E, Malikova I, Confort-
204
Gouny S, Cozzone PJ, Pelletier J, Ranjeva J-P, Audoin B (2012): Assessing brain
connectivity at rest is clinically relevant in early multiple sclerosis. Mult Scler J 18:1251–
1258.
Filippi M, Rocca M a, Benedict RHB, DeLuca J, Geurts JJG, Rombouts S a RB, Ron M, Comi G
(2010): The contribution of MRI in assessing cognitive impairment in multiple sclerosis.
Neurology 75:2121–2128.
Fjell AM, Sneve MH, Grydeland H, Storsve AB, Walhovd KB (2017): The Disconnected Brain
and Executive Function Decline in Aging. Cereb Cortex 27:2303–2317.
Fleischer V, Radetz A, Ciolac D, Muthuraman M, Gonzalez-Escamilla G, Zipp F, Groppa S
(2017): Graph Theoretical Framework of Brain Networks in Multiple Sclerosis: A Review
of Concepts. Neuroscience.
Fling BW, Dale ML, Curtze C, Smulders K, Nutt JG, Horak FB (2016): Associations between
mobility, cognition and callosal integrity in people with parkinsonism. NeuroImage Clin
11:415–422.
Foong J, Rozewicz L, Quaghebeur G, Davie CA, Kartsounis LD, Thompson AJ, Miller DH, Ron
MA (1997): Executive function in multiple sclerosis. The role of frontal lobe pathology.
Brain 120:15–26.
Forn C, Belenguer A, Belloch V, Sanjuan A, Parcet MA, Avila C (2011): Anatomical and
functional differences between the Paced Auditory Serial Addition Test and the Symbol
Digit Modalities Test. J Clin Exp Neuropsychol 33:42–50.
Fornito A, Zalesky A, Breakspear M (2013): Graph analysis of the human connectome: Promise,
progress, and pitfalls. Neuroimage 80:426–444.
Fornito A, Zalesky A, Breakspear M (2015): The connectomics of brain disorders. Nat Rev
205
Neurosci 16:159–172.
Fornito A, Zalesky A, Bullmore E (2016): Fundamentals of Brain Network analysis. Elsevier
Inc.
Friston KJ (1994): Functional and effective connectivity in neuroimaging: A synthesis. Hum
Brain Mapp 2:56–78.
Friston KJ (2011): Functional and effective connectivity: a review. Brain Connect 1:13–36.
Gallagher C, Bell B, Bendlin B, Palotti M, Okonkwo O, Sodhi A, Wong R, Buyan-Dent L,
Johnson S, Willette A, Wilette A, Harding S, Ninman N, Kastman E, Alexander A (2013):
White matter microstructural integrity and executive function in Parkinson’s disease. J Int
Neuropsychol Soc 19:349–54.
Gamboa OL, Tagliazucchi E, Von Wegner F, Jurcoane A, Wahl M, Laufs H, Ziemann U (2014):
Working memory performance of early MS patients correlates inversely with modularity
increases in resting state functional connectivity networks. Neuroimage 94:385–395.
Geisseler O, Pflugshaupt T, Bezzola L, Reuter K, Weller D, Schuknecht B, Brugger P,
Linnebank M (2016): Cortical thinning in the anterior cingulate cortex predicts multiple
sclerosis patients’ fluency performance in a lateralised manner. NeuroImage Clin 10:89–95.
Gelfand JM (2014): Multiple sclerosis: diagnosis, differential diagnosis, and clinical
presentation. In: . Handbook of Clinical Neurology. Elsevier. pp 269–290.
Goldenberg MM (2012): Multiple sclerosis review. Pharm Ther 37:175–184.
Goldman JG, Litvan I (2011): Mild cognitive impairment in Parkinson’s disease. Minerva Med
102:441–459.
Gong G, He Y, Evans AC (2011): Brain connectivity: Gender makes a difference. Neuroscientist
17:575–591.
206
Göttlich M, Münte TF, Heldmann M, Kasten M, Hagenah J, Krämer UM (2013): Altered
Resting State Brain Networks in Parkinson’s Disease. PLoS One 8.
Green PS, Simpkins JW (2000): Neuroprotective effects of estrogens: Potential mechanisms of
action. Int J Dev Neurosci 18:347–358.
Greicius MD, Krasnow B, Reiss AL, Menon V (2003): Functional connectivity in the resting
brain: a network analysis of the default mode hypothesis. Proc Natl Acad Sci U S A
100:253–8.
Guimarães J, Sá MJ (2012): Cognitive dysfunction in Multiple Sclerosis. Front Neurol MAY:1–
8.
Hackmack K, Weygandt M, Wuerfel J, Pfueller CF, Bellmann-Strobl J, Paul F, Haynes JD
(2012): Can we overcome the “clinico-radiological paradox” in multiple sclerosis? J Neurol
259:2151–2160.
Hagmann P, Cammoun L, Gigandet X, Meuli R, Honey C, Wedeen V, Sporns O (2008):
Mapping the structural core of human cerebral cortex. PLoS Biol 6:e159.
Hampson M, Driesen NR, Skudlarski P, Gore JC, Constable RT (2006): Brain Connectivity
Related to Working Memory Performance. J Neurosci 26:13338–13343.
Han K, Chapman SB, Krawczyk DC (2016): Disrupted intrinsic connectivity among default,
dorsal attention, and frontoparietal control networks in individuals with chronic traumatic
brain injury. J Int Neuropsychol Soc 22:263–279.
Handwerker DA, Roopchansingh V, Gonzalez-Castillo J, Bandettini PA (2012): Periodic
changes in fMRI connectivity. Neuroimage 63:1712–1719.
Hänggi J, Fövenyi L, Liem F, Meyer M, Jäncke L (2014): The hypothesis of neuronal
interconnectivity as a function of brain size-a general organization principle of the human
207
connectome. Front Hum Neurosci 8:915.
Harbo HF, Gold R, Tintoré M (2013): Sex and gender issues in multiple sclerosis. Ther Adv
Neurol Disord 6:237–48.
Hawellek D, Hipp J, Lewis C, Corbetta M, Engel A (2011a): Increased functional connectivity
indicates the severity of cognitive impairment in multiple sclerosis. Proc Natl Acad Sci U S
A:19066–19071.
Hawellek DJ, Hipp JF, Lewis CM, Corbetta M, Engel AK (2011b): Increased functional
connectivity indicates the severity of cognitive impairment in multiple sclerosis. Proc Natl
Acad Sci 108:19066–19071.
Helmich RC, Derikx LC, Bakker M, Scheeringa R, Bloem BR, Toni I (2010): Spatial remapping
of cortico-striatal connectivity in parkinson’s disease. Cereb Cortex 20:1175–1186.
Hermundstad AM, Bassett DS, Brown KS, Aminoff EM, Clewett D, Freeman S, Frithsen A,
Johnson A, Tipper CM, Miller MB, Grafton ST, Carlson JM (2013): Structural foundations
of resting-state and task-based functional connectivity in the human brain. Proc Natl Acad
Sci U S A 110:6169–74.
van den Heuvel M, Sporns O (2011): Rich-club organization of the human connectome. J
Neurosci 31:15775–15786.
van den Heuvel MP, Hulshoff Pol HE (2010): Exploring the brain network: A review on resting-
state fMRI functional connectivity. Eur Neuropsychopharmacol 20:519–534.
van den Heuvel MP, de Lange SC, Zalesky A, Seguin C, Yeo BTT, Schmidt R (2017):
Proportional thresholding in resting-state fMRI functional connectivity networks and
consequences for patient-control connectome studies: Issues and recommendations.
Neuroimage 152:437–449.
208
van den Heuvel MP, Sporns O (2013): Network hubs in the human brain. Trends Cogn Sci
17:683–696.
Heyes C (2012): New thinking: the evolution of human cognition. Philos Trans R Soc B Biol Sci
367:2091–2096.
Hindriks R, Adhikari MH, Murayama Y, Ganzetti M, Mantini D, Logothetis NK, Deco G
(2016): Can sliding-window correlations reveal dynamic functional connectivity in resting-
state fMRI? Neuroimage 127:242–256.
Hirano S, Shinotoh H, Eidelberg D (2012): Functional brain imaging of cognitive dysfunction in
Parkinson’s disease. J Neurol Neurosurg Psychiatry 83:963–969.
Hofman MA (2014): Evolution of the human brain: when bigger is better. Front Neuroanat 8:15.
Holland AA, Graves D, Greenberg BM, Harder LL (2014): Fatigue, emotional functioning, and
executive dysfunction in pediatric multiple sclerosis. Child Neuropsychol 20:71–85.
Houtchens MK, Benedict RHB, Killiany R, Sharma J, Jaisani Z, Singh B, Weinstock-Guttman B,
Guttmann CRG, Bakshi R (2007): Thalamic atrophy and cognition in multiple sclerosis.
Neurology 69:1213–1223.
Hutchison RM, Womelsdorf T, Allen EA, Bandettini PA, Calhoun VD, Corbetta M, Della Penna
S, Duyn JH, Glover GH, Gonzalez-Castillo J, Handwerker DA, Keilholz S, Kiviniemi V,
Leopold DA, de Pasquale F, Sporns O, Walter M, Chang C (2013a): Dynamic functional
connectivity: Promise, issues, and interpretations. Neuroimage 80:360–378.
Hutchison RM, Womelsdorf T, Gati JS, Everling S, Menon RS (2013b): Resting-state networks
show dynamic functional connectivity in awake humans and anesthetized macaques. Hum
Brain Mapp 34:2154–2177.
Janssen A, Boster A, Patterson B, Abduljalil A, Prakash R (2013): Resting-state functional
209
connectivity in multiple sclerosis: an examination of group differences and individual
differences. Neuropsychologia 51:2918–2929.
Jones DT, Vemuri P, Murphy MC, Gunter JL, Senjem ML, Machulda MM, Przybelski SA,
Gregg BE, Kantarci K, Knopman DS, Boeve BF, Petersen RC, Jack CR (2012): Non-
stationarity in the “resting brain’s” modular architecture. PLoS One 7.
Julayanont P, Nasreddine ZS (2017): Montreal Cognitive Assessment (MoCA): Concept and
Clinical Review. In: Larner, AJ, editor. Cognitive Screening Instruments: A Practical
Approach. Cham: Springer International Publishing. pp 139–195.
Kaiser RH, Whitfield-Gabrieli S, Dillon DG, Goer F, Beltzer M, Minkel J, Smoski M, Dichter G,
Pizzagalli DA (2015): Dynamic Resting-State Functional Connectivity in Major
Depression. Neuropsychopharmacology:1–9.
Karadayi H, Arisoy O, Altunrende B, Boztas MH, Sercan M (2014): The relationship of
cognitive impairment with neurological and psychiatric variables in multiple sclerosis
patients. Int J Psychiatry Clin Pract 18:45–51.
Kehagia AA, Barker RA, Robbins TW (2010): Neuropsychological and clinical heterogeneity of
cognitive impairment and dementia in patients with Parkinson’s disease. Lancet Neurol
9:1200–1213.
Kehagia AA, Barker RA, Robbins TW (2012): Cognitive impairment in Parkinson’s disease: The
dual syndrome hypothesis. Neurodegener Dis 11:79–92.
Khan S, Gramfort A, Shetty NR, Kitzbichler MG, Ganesan S, Moran JM, Lee SM, Gabrieli JDE,
Tager-Flusberg HB, Joseph RM, Herbert MR, Hamalainen MS, Kenet T (2013): Local and
long-range functional connectivity is reduced in concert in autism spectrum disorders. Proc
Natl Acad Sci 110:3107–3112.
210
Kim J, Criaud M, Cho SS, Díez-Cirarda M, Mihaescu A, Coakeley S, Ghadery C, Valli M,
Jacobs MF, Houle S, Strafella AP (2017): Abnormal intrinsic brain functional network
dynamics in Parkinson’s disease. Brain 140:2955–2967.
Kingwell K (2012): Disease mechanisms in MS: Neuronal network connectivity is altered in
multiple sclerosis. Nat Rev Neurol 8:593.
Knutson B, Rick S, Wimmer GE, Prelec D, Loewenstein G (2007): Neural Predictors of
Purchases. Neuron 53:147–156.
Koshimori Y, Cho S-S, Criaud M, Christopher L, Jacobs M, Ghadery C, Coakeley S, Harris M,
Mizrahi R, Hamani C, Lang AE, Houle S, Strafella AP (2016): Disrupted Nodal and Hub
Organization Account for Brain Network Abnormalities in Parkinson’s Disease. Front
Aging Neurosci 8:1–10.
Kreutzer J, DeLuca J, Caplan B (2011): F-A-S Test. Ed. Jeffrey S. Kreutzer, John DeLuca, Bruce
Caplan. Encyclopedia of Clinical Neuropsychology. Springer New York.
Krzanowski WJ (1988): Principles of multivariate analysis: a user’s perspective. New York:
Clarendon Press.
Kucyi A, Hove MJ, Esterman M, Hutchison RM, Valera EM (2016): Dynamic Brain Network
Correlates of Spontaneous Fluctuations in Attention. Cereb Cortex 27:1831–1840.
Kurtzke J (1983): Rating neurologic impairment in multiple sclerosis: an expanded disability
status scale (EDSS). Neurology 33:1444–52.
Kwak Y, Peltier SJ, Bohnen NI, Müller MLTM, Dayalu P, Seidler RD (2012): L-DOPA changes
spontaneous low-frequency BOLD signal oscillations in Parkinson’s disease: a resting state
fMRI study. Front Syst Neurosci 6:52.
Langdon D, Amato M, Boringa J, Brochet B, Foley F, Fredrikson S, Hamalainen P, Hartung H-
211
P, Krupp L, Penner I, Reder A, Benedict R (2012): Recommendations for a Brief
International Cognitive Assessment for Multiple Sclerosis (BICAMS). Mult Scler J 18:891–
898.
Langdon DW (2011): Cognition in multiple sclerosis. Curr Opin Neurol 24:244–249.
Lange F, Seer C, Kopp B (2017): Cognitive flexibility in neurological disorders: Cognitive
components and event-related potentials. Neurosci Biobehav Rev 83:496–507.
Leavitt VM, Wylie G, Krch D, Chiaravalloti N, DeLuca J, Sumowski JF (2014): Does slowed
processing speed account for executive deficits in multiple sclerosis? Evidence from
neuropsychological performance and structural neuroimaging. Rehabil Psychol 59:422–428.
Lebedev A V., Westman E, Simmons A, Lebedeva A, Siepel FJ, Pereira JB, Aarsland D (2014):
Large-scale resting state network correlates of cognitive impairment in Parkinson’s disease
and related dopaminergic deficits. Front Syst Neurosci 8:1–12.
Lee D, Fletcher E, Martinez O, Zozulya N, Kim J, Tran J, Buonocore M, Carmichael O, DeCarli
C (2010): Vascular and degenerative processes differentially affect regional
interhemispheric connections in normal aging, mild cognitive impairment, and Alzheimer
disease. Stroke 41:1791–1797.
Leonardi N, Richiardi J, Gschwind M, Simioni S, Annoni JM, Schluep M, Vuilleumier P, Van
De Ville D (2013): Principal components of functional connectivity: A new approach to
study dynamic brain connectivity during rest. Neuroimage 83:937–950.
Leonardi N, Van De Ville D (2015): On spurious and real fluctuations of dynamic functional
connectivity during rest. Neuroimage 104:430–436.
Liao X, Yuan L, Zhao T, Dai Z, Shu N, Xia M, Yang Y, Evans A, He Y (2015): Spontaneous
functional network dynamics and associated structural substrates in the human brain. Front
212
Hum Neurosci 9:478.
Lin S-J, Lam J, Beveridge S, Vavasour I, Traboulsee A, Li DKB, MacKay A, McKeown M,
Kosaka B (2017): Cognitive Performance in Subjects With Multiple Sclerosis Is Robustly
Influenced by Gender in Canonical-Correlation Analysis. J Neuropsychiatry Clin Neurosci
29:119–127.
Lindquist MA (2008): The Statistical Analysis of fMRI Data. Stat Sci 23:439–464.
Liu Y, Wang H, Duan Y, Huang J, Ren Z, Ye J, Dong H, Shi F, Li K, Wang J (2017): Functional
Brain Network Alterations in Clinically Isolated Syndrome and Multiple Sclerosis: A
Graph-based Connectome Study. Radiology 282:534–541.
Llufriu S, Martinez-Heras E, Solana E, Sola-Valls N, Sepulveda M, Blanco Y, Martinez-
Lapiscina EH, Andorra M, Villoslada P, Prats-Galino A, Saiz A (2017): Structural networks
involved in attentional function in multiple sclerosis. Neuroimage 13:288–296.
Loitfelder M, Filippi M, Rocca M, Valsasina P, Ropele S, Jehna M, Fuchs S, Schmidt R, Neuper
C, Fazekas F, Enzinger C (2012): Abnormalities of resting state functional connectivity are
related to sustained attention deficits in MS. PLoS One 7:1–9.
Lord A, Horn D, Breakspear M, Walter M (2012): Changes in community structure of resting
state functional connectivity in unipolar depression. PLoS One 7.
Louapre C, Perlbarg V, García-Lorenzo D, Urbanski M, Benali H, Assouad R, Galanaud D,
Freeman L, Bodini B, Papeix C, Tourbah A, Lubetzki C, Lehéricy S, Stankoff B (2014):
Brain networks disconnection in early multiple sclerosis cognitive deficits: An
anatomofunctional study. Hum Brain Mapp 35:4706–4717.
Luo CY, Guo XY, Song W, Chen Q, Cao B, Yang J, Gong QY, Shang HF (2015a): Functional
connectome assessed using graph theory in drug-naive Parkinson’s disease. J Neurol
213
262:1557–1567.
Luo C, Guo X, Song W, Zhao B, Cao B, Yang J, Gong Q, Shang HF (2015b): Decreased resting-
state interhemispheric functional connectivity in Parkinson’s disease. Biomed Res Int
2015:8 pages.
Luo C, Song W, Chen Q, Zheng Z, Chen K, Cao B, Yang J, Li J, Huang X, Gong Q, Shang HF
(2014): Reduced functional connectivity in early-stage drug-naive Parkinson’s disease: A
resting-state fMRI study. Neurobiol Aging 35:431–441.
Madhyastha TM, Askren MK, Boord P, Grabowski TJ (2014): Dynamic connectivity at rest
predicts attention task performance. Brain Connect 5:45–59.
Margulies DS, Böttger J, Long X, Lv Y, Kelly C, Schäfer A, Goldhahn D, Abbushi A, Milham
MP, Lohmann G, Villringer A (2010): Resting developments: A review of fMRI post-
processing methodologies for spontaneous brain activity. Magn Reson Mater Physics, Biol
Med 23:289–307.
Marrie RA (2016): Comorbidity in multiple sclerosis: Some answers, more questions. Int J MS
Care 18:271–272.
Marrie RA (2017): Comorbidity in multiple sclerosis: Implications for patient care. Nat Rev
Neurol 13:375–382.
Mattar MG, Cole MW, Thompson-Schill SL, Bassett DS (2015): A Functional Cartography of
Cognitive Systems. PLoS Comput Biol 11:1–26.
Mattila PM, Röyttä M, Lönnberg P, Marjamäki P, Helenius H, Rinne JO (2001): Choline
acetytransferase activity and striatal dopamine receptors in Parkinson’s disease in relation to
cognitive impairment. Acta Neuropathol 102:160–6.
Mavandadi S, Nazem S, Ten Have TR, Siderowf AD, Duda JE, Stern MB, Weintraub D (2009):
214
Use of latent variable modeling to delineate psychiatric and cognitive profiles in parkinson
disease. Am J Geriatr Psychiatry 17:986–995.
Mayka MA, Corcos DM, Leurgans SE, Vaillancourt DE (2006): Three-dimensional locations
and boundaries of motor and premotor cortices as defined by functional brain imaging: A
meta-analysis. Neuroimage 31:1453–1474.
McCabe DP, Roediger III HL, McDaniel MA, Balota DA, Hambrick DZ (2010): The
Relationship Between Working Memory Capacity and Executive Functioning: Evidence for
a Common Executive Attention Construct. Neuropsychology 24:222–243.
McIntosh AR (2000): Towards a network theory of cognition. Neural Networks 13:861–870.
McIntosh AR (2004): Contexts and catalysts: a resolution of the localization and integration of
function in the brain. Neuroinformatics 2:175–182.
McIntosh AR, Kovacevic N, Itier RJ (2008): Increased brain signal variability accompanies
lower behavioral variability in development. PLoS Comput Biol 4.
McKeown MJ, Hansen LK, Sejnowski T (2003): Independent component analysis of functional
MRI: what is signal and what is noise? Curr Opin Neurobiol 13:620–9.
McNamara P, Durso R (2018): The dopamine system, Parkinson’s disease and language
function. Curr Opin Behav Sci 21:1–5.
Meijer KA, Eijlers AJC, Douw L, Uitdehaag BMJ, Barkhof F, Geurts, J.J.G. (2017): Increased
connectivity of hub networks and cognitive impairment in multiple sclerosis. Neurology
88:2107–2114.
Menza MA, Robertson-Hoffman DE, Bonapace AS (1993): Parkinson’s disease and anxiety:
Comorbidity with depression. Biol Psychiatry 34:465–470.
Miller DI, Halpern DF (2014): The new science of cognitive sex differences. Trends Cogn Sci
215
18:37–45.
Miller EK, Cohen JD (2001): An Integrative Theory of Prefrontl Cortex Function. Annu Rev
Neurosci 24:167–202.
Miller EK, Wallis JD (2009): Executive Function and Higher-Order Cognition : Definition and
Neural Substrates. In: . Encyclopedia of Neuroscience. Elsevier. Vol. 4, pp 99–104.
Miller IN, Cronin-Golomb A (2010): Gender differences in Parkinson’s disease: Clinical
characteristics and cognition. Mov Disord 25:2695–2703.
Miller IN, Neargarder S, Risi MM, Cronin-Golomb A (2013): Frontal and posterior subtypes of
neuropsychological deficit in Parkinson’s disease. Behav Neurosci 127:175–83.
Mitolo M, Venneri A, Wilkinson ID, Sharrack B (2015): Cognitive rehabilitation in multiple
sclerosis: A systematic review. J Neurol Sci 354:1–9.
Möller C, Hafkemeijer A, Pijnenburg YAL, Rombouts SARB, van der Grond J, Dopper E, van
Swieten J, Versteeg A, Steenwijk MD, Barkhof F, Scheltens P, Vrenken H, van der Flier
WM (2016): Different patterns of cortical gray matter loss over time in behavioral variant
frontotemporal dementia and Alzheimer’s disease. Neurobiol Aging 38:21–31.
Monchi O, Petrides M, Strafella AP, Worsley KJ, Doyon J (2006): Functional role of the basal
ganglia in the planning and execution of actions. Ann Neurol 59:257–264.
Murman DL (2015): The Impact of Age on Cognition. Semin Hear 36:111–121.
Mwansisya T, Wang Z, Tao H, Zhang H, Hu A, Guo S, Liu Z (2013): The diminished
interhemispheric connectivity correlates with negative symptoms and cognitive impairment
in first-episode schizophrenia. Schizophr Res 150:144–150.
Narayanan NS, Rodnitzky RL, Uc E (2013): Prefrontal dopamine signaling and cognitive
symptoms of Parkinson’s Disease. Rev Neurosci 24:267–78.
216
Nilsson MH, Ullén S, Ekström H, Iwarsson S (2016): The association between indicators of
health and housing in people with Parkinson’s disease. BMC Geriatr 16:146.
Nomi JS, Vij SG, Dajani DR, Steimke R, Damaraju E, Rachakonda S, Calhoun VD, Uddin LQ
(2017): Chronnectomic patterns and neural flexibility underlie executive function.
Neuroimage 147:861–871.
Olde KTE, Deijen JB, Barkhof F (2014): Functional connectivity and cognitive decline over 3
years in Parkinson disease. Neurology 83:2046–53.
Ouchi Y, Kanno T, Okada H, Yoshikawa E, Futatsubashi M, Nobezawa S, Torizuka T, Tanaka K
(2001): Changes in dopamine availability in the nigrostriatal and mesocortical dopaminergic
systems by gait in Parkinson’s disease. Brain 124:784–792.
Owen AM (2004): Cognitive dysfunction in Parkinson’s disease: the role of frontostriatal
circuitry. Neuroscientist 10:525–537.
Pagonabarraga J, Kulisevsky J (2012): Cognitive impairment and dementia in Parkinson’s
disease. Neurobiol Dis 46:590–596.
Palavra NC, Naismith SL, Lewis SJ (2013): Mild cognitive impairment in Parkinson’s disease: a
review of current concepts. Neurol Res Int 2013:576091.
Palmer MW (1993): Putting things in even better order: The advantages of Canonical correlation
analysis. Ecology 74:2215–2230.
Pamplona GSP, Santos Neto GSGS, Rosset SRE, Rogers BP, Salmon CEG (2015): Analyzing
the association between functional connectivity of the brain and intellectual performance.
Front Hum Neurosci 9:1–11.
Park HJ, Friston K (2013): Structural and functional brain networks: From connections to
cognition. Science (80- ) 342.
217
Parker KL, Lamichhane D, Caetano MS, Narayanan NS (2013): Executive dysfunction in
Parkinson’s disease and timing deficits. Front Integr Neurosci 7:75.
Parlatini V, Radua J, Dell’Acqua F, Leslie A, Simmons A, Murphy DG, Catani M, Thiebaut de
Schotten M (2016): Functional segregation and integration within fronto-parietal networks.
Neuroimage 146:367–375.
Parmenter BA, Weinstock-Guttman B, Garg N, Munschauer F, Benedict RH (2007): Screening
for cognitive impairment in multiple sclerosis using the Symbol Digit Modalities Test. Mult
Scler J 13:52–57.
Paul L, Brown W, Adolphs R, Tyszka J, Richards L, Mukherjee P, Sherr E (2007): Agenesis of
the corpus callosum: genetic, developmental and functional aspects of connectivity. Nat
Rev Neurosci 8:287–299.
Perry A, Wen W, Kochan NA, Thalamuthu A, Sachdev PS, Breakspear M (2017): The
independent influences of age and education on functional brain networks and cognition in
healthy older adults. Hum Brain Mapp 38:5094–5114.
Petersen SE, Sporns O (2015): Brain Networks and Cognitive Architectures. Neuron 88:207–
219.
Pievani M, Filippini N, van den Heuvel MP, Cappa SF, Frisoni GB (2014): Brain connectivity in
neurodegenerative diseases—from phenotype to proteinopathy. Curr Opin Neurobiol 10:1–
14.
Poewe W (2008): Non-motor symptoms in Parkinson’s disease. Eur J Neurol 15:14–20.
Polak PE, Kalinin S, Feinstein DL (2011): Locus coeruleus damage and noradrenaline reductions
in multiple sclerosis and experimental autoimmune encephalomyelitis. Brain 134:665–677.
Polman C, Reingold S, Edan G, Filippi M, Hartung H, Kappos L, Lublin F, Metz L, McFarland
218
H, O’Connor P, Sandberg-Wollheim M, Thompson A, Weinshenker B, Wolinsky J (2005):
Diagnostic criteria for multiple sclerosis: 2005 revisions to the McDonald criteria. Ann
Neurol 58:840–846.
Power JD, Schlaggar BL, Lessov-Schlaggar CN, Petersen SE (2013): Evidence for hubs in
human functional brain networks. Neuron 79:798–813.
Preston J, Hammersley R, Gallagher H (2013): The executive dysfunctions most commonly
associated with multiple sclerosis and their impact on occupational performance. Br J
Occup Ther 76:225–233.
Preti MG, Bolton TA, Van De Ville D (2017): The dynamic functional connectome: State-of-
the-art and perspectives. Neuroimage 160:41–54.
Prodoehl J, Burciu RG, Vaillancourt DE (2014): Resting state functional magnetic resonance
imaging in Parkinson’s disease. Curr Neurol Neurosci Rep 14.
Putcha D, Ross RS, Cronin-Golomb A, Janes AC, Stern CE (2015): Altered intrinsic functional
coupling between core neurocognitive networks in Parkinson’s disease. NeuroImage Clin
7:449–455.
Rahn K, Slusher B, Kaplin A (2012): Cognitive impairment in multiple sclerosis: a forgotten
disability remembered. Cerebrum 2012.
Reineberg AE, Andrews-Hanna JR, Depue BE, Friedman NP, Banich MT (2015): Resting-state
networks predict individual differences in common and specific aspects of executive
function. Neuroimage 104:69–78.
Reineberg AE, Banich MT (2016): Functional connectivity at rest is sensitive to individual
differences in executive function: A network analysis. Hum Brain Mapp 37:2959–2975.
Richiardi J, Gschwind M, Simioni S, Annoni J-M, Greco B, Hagmann P, Schluep M,
219
Vuilleumier P, Van De Ville D (2012): Classifying minimally disabled multiple sclerosis
patients from resting state functional connectivity. Neuroimage 62:2021–33.
Riedel O, Klotsche J, Spottke A, Deuschl G, Förstl H, Henn F, Heuser I, Oertel W, Reichmann
H, Riederer P, Trenkwalder C, Dodel R, Wittchen HU (2008): Cognitive impairment in 873
patients with idiopathic Parkinson’s disease: Results from the German Study on
Epidemiology of Parkinson’s Disease with Dementia (GEPAD). J Neurol 255:255–264.
Rocca MR, De Meo E, Amato MP, Copetti M, Moiola L, Ghezzi A, Veggiotti P, Capra R,
Fiorino A, Pippolo L, Pera MC, Falini A, Comi G, Filippi M (2014): Cognitive impairment
in paediatric multiple sclerosis patients is not related to cortical lesions. Mult Scler J:1–4.
Rocca MA, Valsasina P, Martinelli V, Misci P, Falini A, Comi G, Filippi M (2012): Large-scale
neuronal network dysfunction in relapsing-remitting multiple sclerosis. Neurology:1449–
1457.
Rocca M a., Valsasina P, Meani A, Falini A, Comi G, Filippi M (2016a): Impaired functional
integration in multiple sclerosis: a graph theory study. Brain Struct Funct 221:115–131.
Rocca M, Valsasina P, Meani A, Falini A, Comi G, Filippi M (2016b): Impaired functional
integration in multiple sclerosis: a graph theory study. Brain Struct Funct 221:115–131.
Rosazza C, Minati L (2011): Resting-state brain networks: Literature review and clinical
applications. Neurol Sci 32:773–785.
Rossi F, Giorgio A, Battaglini M, Stromillo ML, Portaccio E, Goretti B, Federico A, Hakiki B,
Amato MP, de Stefano N (2012): Relevance of Brain Lesion Location to Cognition in
Relapsing Multiple Sclerosis. PLoS One 7:1–7.
Rubinov M, Sporns O (2010): Complex network measures of brain connectivity: Uses and
interpretations. Neuroimage 52:1059–1069.
220
Saar-Ashkenazy R, Cohen J, Guez J, Gasho C, Shelef I, Friedman A, Shalev H (2014): Reduced
corpus-callosum volume in posttraumatic stress disorder highlights the importance of
interhemispheric connectivity for associative memory. J Trauma Stress 27:481–488.
Saenger VM, Kahan J, Foltynie T, Friston K, Aziz TZ, Green AL, Van Hartevelt TJ, Cabral J,
Stevner ABA, Fernandes HM, Mancini L, Thornton J, Yousry T, Limousin P, Zrinzo L,
Hariz M, Marques P, Sousa N, Kringelbach ML, Deco G (2017): Uncovering the
underlying mechanisms and whole-brain dynamics of deep brain stimulation for
Parkinson’s disease. Sci Rep 7:1–14.
Sakoglu U, Pearlson GD, Kiehl KA, Wang YM, Michael AM, Calhoun VD (2010): A method
for evaluating dynamic functional network connectivity and task-modulation: Application
to schizophrenia. Magn Reson Mater Physics, Biol Med 23:351–366.
Salgado S, Kaplitt MG (2015): The nucleus accumbens: A comprehensive review. Stereotact
Funct Neurosurg 93:75–93.
Salmelin R, Kujala J (2006): Neural representation of language: activation versus long-range
connectivity. Trends Cogn Sci 10:519–525.
Santarnecchi E, Galli G, Polizzotto NR, Rossi A, Rossi S (2014): Efficiency of weak brain
connections support general cognitive functioning. Hum Brain Mapp 35:4566–4582.
Sasson E, Doniger GM, Pasternak O, Tarrasch R, Assaf Y (2013): White matter correlates of
cognitive domains in normal aging with diffusion tensor imaging. Front Neurosci 7:1–13.
Savettieri G, Messina D, Andreoli V, Bonavita S, Caltagirone C, Cittadella R, Farina D, Fazio
MC, Girlanda P, Le Pira F, Liguori M, Lugaresi A, Nocentini U, Reggio A, Salemi G,
Tedeschi G, Trojano M, Valentino P, Quattrone A (2004): Gender-related effect of clinical
and genetic variables on the cognitive impairment in multiple sclerosis. J Neurol 251:1208–
221
1214.
Sawamoto N, Honda M, Hanakawa T, Fukuyama H, Shibasaki H (2002): Cognitive slowing in
Parkinson’s disease: a behavioral evaluation independent of motor slowing. J Neurosci
22:5198–5203.
Scheller E, Minkova L, Leitner M, Klöppel S (2014): Attempted and successful compensation in
preclinical and early manifest neurodegeneration - a review of task fMRI studies. Front
Psychiatry 5:1–16.
Schoonheim MM, Hulst HE, Landi D, Ciccarelli O, Roosendaal SD, Sanz-Arigita EJ, Vrenken
H, Polman CH, Stam CJ, Barkhof F, Geurts JJG (2012a): Gender-related differences in
functional connectivity in multiple sclerosis. Mult Scler J 18:164–173.
Schoonheim MM, Meijer K a., Geurts JJG (2015): Network Collapse and Cognitive Impairment
in Multiple Sclerosis. Front Neurol 6:1–5.
Schoonheim MM, Popescu V, Lopes FCR, Wiebenga OT, Vrenken H, Douw L, Polman CH,
Geurts JJG, Barkhof F (2012b): Subcortical atrophy and cognition: Sex effects in multiple
sclerosis. Neurology 79:1754–1761.
Schoonheim MM, Vigeveno RM, Lopes FCR, Pouwels PJW, Polman CH, Barkhof F, Geurts
JJG (2014): Sex-specific extent and severity of white matter damage in multiple sclerosis:
Implications for cognitive decline. Hum Brain Mapp 35:2348–2358.
Schoonheim M, Geurts J, Wiebenga O, De Munck J, Polman C, Stam C, Barkhof F, Wink A
(2013): Changes in functional network centrality underlie cognitive dysfunction and
physical disability in multiple sclerosis. Mult Scler 20:1058–1065.
Scimeca JM, Badre D (2012): Striatal Contributions to Declarative Memory Retrieval. Neuron
75:380–392.
222
Seeley WW, Menon V, Schatzberg AF, Keller J, Glover GH, Kenna H, Reiss AL, Greicius MD
(2007): Dissociable intrinsic connectivity networks for salience processing and executive
control. J Neurosci 27:2349–2356.
Seibert TM, Murphy EA, Kaestner EJ, Brewer JB (2012): Interregional Correlations in
Parkinson Disease and Parkinson-related Dementia with Resting Functional MR Imaging.
Radiology 263:226–234.
Shafto MA, Tyler LK (2014): Language in the aging brain: The network dynamics of cognitive
decline and preservation. Science (80- ) 346:583–588.
Shah LM, Cramer JA, Ferguson MA, Birn RM, Anderson JS (2016): Reliability and
reproducibility of individual differences in functional connectivity acquired during task and
resting state. Brain Behav 6:1–15.
Sharman M, Valabregue R, Perlbarg V, Marrakchi-Kacem L, Vidailhet M, Benali H, Brice A,
Lehéricy S (2013): Parkinson’s disease patients show reduced cortical-subcortical
sensorimotor connectivity. Mov Disord 28:447–454.
Shaw EE, Schultz AP, Sperling RA, Hedden T (2015): Functional Connectivity in Multiple
Cortical Networks Is Associated with Performance Across Cognitive Domains in Older
Adults. Brain Connect 5:505–16.
Sheffield JM, Repovs G, Harms MP, Carter CS, Gold JM, MacDonald AW, Daniel Ragland J,
Silverstein SM, Godwin D, Barch DM (2015): Fronto-parietal and cingulo-opercular
network integrity and cognition in health and schizophrenia. Neuropsychologia 73:82–93.
Sheridan LK, Fitzgerald HE, Adams KM, Nigg JT, Martel MM, Puttler LI, Wong MM, Zucker
RA (2006): Normative Symbol Digit Modalities Test performance in a community-based
sample. Arch Clin Neuropsychol 21:23–28.
223
Sherwin BB (2012): Estrogen and Cognitive Functioning in Women: Lessons We Have Learned.
Behav Neurosci 126:123–127.
Shine JM, Bissett PG, Bell PT, Koyejo O, Balsters JH, Gorgolewski KJ, Moodie CA, Poldrack
RA (2016): The Dynamics of Functional Brain Networks: Integrated Network States during
Cognitive Task Performance. Neuron 92:544–554.
Shu N, Duan Y, Xia M, Schoonheim MM, Huang J, Ren Z, Sun Z, Ye J, Dong H, Shi FD,
Barkhof F, Li K, Liu Y (2016): Disrupted topological organization of structural and
functional brain connectomes in clinically isolated syndrome and multiple sclerosis. Sci Rep
6:1–11.
Siegel JS, Ramsey LE, Snyder AZ, Metcalf N V., Chacko R V., Weinberger K, Baldassarre A,
Hacker CD, Shulman GL, Corbetta M (2016): Disruptions of network connectivity predict
impairment in multiple behavioral domains after stroke. Proc Natl Acad Sci 113:E4367–
E4376.
Siepel FJ, Brønnick KS, Booij J, Ravina BM, Lebedev A V., Pereira JB, Grüner R, Aarsland D
(2014): Cognitive executive impairment and dopaminergic deficits in de novo Parkinson’s
disease. Mov Disord 29:1802–1808.
Simioni AC, Dagher A, Fellows LK (2016): Compensatory striatal-cerebellar connectivity in
mild-moderate Parkinson’s disease. NeuroImage Clin 10:54–62.
Smith S (2015): Linking cognition to brain connectivity. Nat Neurosci 19:7–9.
Smith SM, Vidaurre D, Beckmann CF, Glasser MF, Jenkinson M, Miller KL, Nichols TE,
Robinson EC, Salimi-Khorshidi G, Woolrich MW, Barch DM, Uǧurbil K, Van Essen DC
(2013): Functional connectomics from resting-state fMRI. Trends Cogn Sci 17:666–682.
Smith SM, Fox PT, Miller KL, Glahn DC, Fox PM, Mackay CE, Filippini N, Watkins KE, Toro
224
R, Laird AR, Beckmann CF (2009): Correspondence of the brain’s functional architecture
during activation and rest. Proc Natl Acad Sci U S A 106:13040–5.
Smith SM, Nichols TE, Vidaurre D, Winkler AM, Behrens TEJ, Glasser MF, Ugurbil K, Barch
DM, Van Essen DC, Miller KL (2015): A positive-negative mode of population covariation
links brain connectivity, demographics and behavior. Nat Neurosci 18:1565–1567.
Sonder J, Burggraaff J, Knol D, Polman C, Uitdehaag B (2014): Comparing long-term results of
PASAT and SDMT scores in relation to neuropsychological testing in multiple sclerosis.
Mult Scler 20:481–488.
Specogna I, Casagrande F, Lorusso A, Catalan M, Gorian A, Zugna L, Longo R, Zorzon M,
Naccarato M, Pizzolato G, Ukmar M, Cova M (2012): Functional MRI during the execution
of a motor task in patients with multiple sclerosis and fatigue. Radiol Med 117:1389–1407.
Sporns O (2013): Network attributes for segregation and integration in the human brain. Curr
Opin Neurobiol 23:162–171.
Spreng RN, Sepulcre J, Turner GR, Stevens WD, Schacter DL (2012): Intrinsic Architecture
Underlying the Relations among the Default, Dorsal Attention, and Fronto-parietal Control
Networks of the Human Brain. J Cogn Neurosci:1–12.
Sridharan D, Levitin DJ, Menon V (2008): A critical role for the right fronto-insular cortex in
switching between central-executive and default-mode networks. Proc Natl Acad Sci U S A
105:12569–12574.
Stam CJ (2014): Modern network science of neurological disorders. Nat Rev Neurosci 15:683–
695.
Stankiewicz JM, Glanz BI, Healy BC, Arora A, Neema M, Benedict RHB, Guss ZD, Tauhid S,
Buckle GJ, Houtchens MK, Khoury SJ, Weiner HL, Guttmann CRG, Bakshi R (2009):
225
Brain MRI Lesion Load at 1 . 5T and 3T versus Clinical Status in Multiple Sclerosis. J
Neuroimhaing 21:50–56.
Starkstein SE, Preziosi TJ, Berthier ML, Bolduc PL, Mayberg HS, Robinson RG (1989):
Depression and cognitive impairment in Parkinson’s Disease. Brain 112:1141–1153.
Stephens GJ, Honey CJ, Hasson U (2013): A place for time: the spatiotemporal structure of
neural dynamics during natural audition. J Neurophysiol 110:2019–26.
Stern Y (2002): What is cognitive reserve? Theory and research application of the reserve
concept. J Int Neuropsychol Soc 8:448–460.
Stoffers D, Bosboom JLW, Deijen JB, Wolters EC, Stam CJ, Berendse HW (2008): Increased
cortico-cortical functional connectivity in early-stage Parkinson’s disease: An MEG study.
Neuroimage 41:212–222.
Strober L, Englert J, Munschauer F, Weinstock-Guttman B, Rao S, Benedict RHB (2009):
Sensitivity of conventional memory tests in multiple sclerosis: comparing the Rao Brief
Repeatable Neuropsychological Battery and the Minimal Assessment of Cognitive Function
in MS. Mult Scler 15:1077–1084.
Sui J, He H, Pearlson GD, Adali T, Kiehl KA, Yu Q, Clark VP, Castro E, White T, Mueller BA,
Ho BC, Andreasen NC, Calhoun VD (2013): Three-way (N-way) fusion of brain imaging
data based on mCCA+jICA and its application to discriminating schizophrenia. Neuroimage
66:119–132.
Takahashi H, Yamada M, Suhara T (2012): Functional significance of central D1 receptors in
cognition: beyond working memory. J Cereb Blood Flow Metab 32:1248–58.
Tam CWC, Burton EJ, McKeith IG, Burn DJ, O’Brien JT (2005): Temporal lobe atrophy on
MRI in Parkinson disease with dementia: a comparison with Alzheimer disease and
226
dementia with Lewy bodies. Neurology 64:861–865.
Telesford QK, Morgan AR, Hayasaka S, Simpson SL, Barret W, Kraft R a, Mozolic JL,
Laurienti PJ (2010): Reproducibility of graph metrics in FMRI networks. Front
Neuroinform 4:117.
Thenganatt MA, Jankovic J (2014): Parkinson disease subtypes. JAMA Neurol 71:499–504.
Thompson GJ, Magnuson ME, Merritt MD, Schwarb H, Pan WJ, Mckinley A, Tripp LD,
Schumacher EH, Keilholz SD (2013): Short-time windows of correlation between large-
scale functional brain networks predict vigilance intraindividually and interindividually.
Hum Brain Mapp 34:3280–3298.
Tucker-Drob E, Johnson K, Jones R (2011): The Cognitive Reserve Hypothesis: a Longitudinal
Examination of Age- Associated Declines in Reasoning and Processing Speed. Dev Psychol
45:431–446.
Vance E, Roberson A, McGunness T, Fazeli P (2010): How Neuroplasticity and Cognitive
Reserve Protect Cognitive Functioning. J Psychosoc Nurs 48:23–30.
Varoquaux G, Thirion B (2014): How machine learning is shaping cognitive neuroimaging.
Gigascience 3:1–7.
Vervoort G, Alaerts K, Bengevoord A, Nackaerts E, Heremans E, Vandenberghe W, Nieuwboer
A (2016): Functional connectivity alterations in the motor and fronto-parietal network relate
to behavioral heterogeneity in Parkinson’s disease. Park Relat Disord 24:48–55.
Wallin MT, Wilken JA, Kane R (2006): Cognitive dysfunction in multiple sclerosis: Assessment,
imaging, and risk factors. J Rehabil Res Dev 43:63.
Wang L, Li K, Zhang Q-E, Zeng Y-W, Jin Z, Dai W-J, Su Y-A, Wang G, Tan Y-L, Yu X, Si T-
M (2013): Interhemispheric functional connectivity and its relationships with clinical
227
characteristics in major depressive disorder: a resting state fMRI study. PLoS One
8:e60191.
Wee CY, Yang S, Yap PT, Shen D (2013): Temporally dynamic resting-state functional
connectivity networks for early MCI identification. Lect Notes Comput Sci (including
Subser Lect Notes Artif Intell Lect Notes Bioinformatics) 8184 LNCS:139–146.
Williams-Gray CH, Foltynie T, Brayne CEG, Robbins TW, Barker RA (2007): Evolution of
cognitive dysfunction in an incident Parkinson’s disease cohort. Brain 130:1787–1798.
Wu T, Long X, Zang Y, Wang L, Hallett M, Li K, Chan P (2009): Regional homogeneity
changes in patients with parkinson’s disease. Hum Brain Mapp 30:1502–1510.
Xia M, Wang J, He Y (2013): BrainNet Viewer: A Network Visualization Tool for Human Brain
Connectomics. PLoS One 8.
Zahodne LB, Glymour MM, Sparks C, Bontempo D, Dixon RA, MacDonald SWS, Manly JJ
(2011): Education Does Not Slow Cognitive Decline with Aging: 12-Year Evidence from
the Victoria Longitudinal Study. J Int Neuropsychol Soc 17:1039–1046.
Zalesky A, Fornito A, Cocchi L, Gollo LL, Breakspear M (2014): Time-resolved resting-state
brain networks. Proc Natl Acad Sci 111:10341–10346.
Zalesky A, Breakspear M (2015): Towards a statistical test for functional connectivity dynamics.
Neuroimage 114:466–470.
Zeighami Y, Fereshtehnejad S-M, Dadar M, Collins DL, Postuma RB, Mišić B, Dagher A
(2017): A clinical-anatomical signature of Parkinson’s disease identified with partial least
squares and magnetic resonance imaging. Neuroimage [Epub ahea.
Zheng Z, Shemmassian S, Wijekoon C, Kim W, Bookheimer SY, Pouratian N (2014): DTI
correlates of distinct cognitive impairments in Parkinson’s disease. Hum Brain Mapp
228
35:1325–1333.
Zhou Y, Milham M, Zuo XN, Kelly C, Jaggi H, Herbert J, Grossman RI, Ge Y (2013):
Functional homotopic changes in multiple sclerosis with resting-state functional MR
imaging. AJNR Am J Neuroradiol 34:1180–1187.
Zhu Y, Song X, Xu M, Hu X, Li E, Liu J, Yuan Y, Gao JH, Liu W (2016): Impaired
interhemispheric synchrony in Parkinson’s disease with depression. Sci Rep 6:1–9.
229
Appendices
Appendix A
Appendix A lists supporting materials of Chapter 2.
A.1 Canonical loadings of individual PD subjects.
The figure illustrates canonical loadings (correlation between transformed CCA data and raw
scores of CCA input) of the cognitive variables which show significant impacts in the CCA model.
In each score, blue dots represent female data and red dots are male data. Average canonical
loadings (indicated as r in the legend) are calculated as well as average cognitive z-scores in both
groups. Except BJLO (the upper right corner), female subjects show higher average cognitive
scores, indicating better performances.
230
Appendix B
Appendix B lists the supporting materials of Chapter 3.
B.1 Robust regression in PD on-medication
Only medication dose could be predicted with the interhemispheric connectivity of the ventral
medial prefrontal cortex in PD on-medication. The rest scores cannot be well-predicted (R-square
< 0.2) by this interhemispheric connectivity pair. The red line indicates perfect prediction. The
blue line indicates current prediction.
231
B.2 Whole brain connectivity matrix of HC and PD
Simple correlation (Pearson’s r) was conducted among 54 ROIs in normal control (NC), PF on-
medication, and PD off-medication subjects. The figure shows the average connectivity matrix of
three populations. Overall, PD off-medication (the right panel) showed higher global connectivity.
Compared to NC, PD off-medication and PD on-medication did not show significant connectivity
difference after corrected for FDR. However, on-medication and off-medication demonstrated
significant different connections. The results were reported in section 3.3.
B.3 Correlation between long-range connections and cognitive performance
In order to study whether the long-range connections observed in MS were related to cognitive
functions, we performed a simple post hoc analysis. Connections significantly different between
HC and MS in figure 3.6 were averaged and pairwise correlation was conducted on the mean long-
range connectivity against cognitive scores in MS.
232
SDMT scores did not show correlation to the mean connectivity values; however, PASAT
performance showed moderate correlation with the mean long-range connection (r=0.42, p=0.036)
as shown in this figure.
The results further support that SDMT performance requires more interhemispheric connectivity,
while PASAT performance engages more long-range connections linking frontal and parietal
areas.
In PD, connections which were significantly different between PD on-medication and off-
medication did not show any correlations to the clinical scores.
233
Appendix C
Appendix C lists the supporting materials of Chapter 4.
C.1 Significantly different local measures between PD and HS
Nine ROIs show higher local efficiency in PD indicated with black stars (p<0.05, uncorrected).
These ROIs are the left amygdala, left middle temporal gyrus, left postcentral gyrus, left angular
gyrus, left supramarginal gyrus, left pre-motor area, right hippocampus, right entorhinal cortex,
and right postcentral gyrus.
234
Seven ROIs show altered betweenness centrality in PD, which are indicated with black starts
(p<0.05, uncorrected). The right pallidum and right accumbens areas show increased betweenness
centrality; while the rest ROIs show decreased values such as the left superior frontal gyrus, left
middle temporal gyrus, left superior parietal gyrus, right inferior frontal gyrus, and right superior
parietal gyrus.
C.2 Logistic LASSO identifies graphical measures into PD and HS groups
Logistic LASSO was further applied to test whether the graphical measures could be used to
distinguish PD from HS, as well as indicating those measures with the largest contribution (Matlab
function lassoglm). With logistic LASSO, the ROIs which showed significant uncorrected p
values in the t-tests on local measures were all important contributors to the separation of PD and
HS (Figure C.3) except betweenness centrality in the right superior parietal gyrus, which was
235
almost insignificant in t-tests (p=0.0415). For global measures, logistic LASSO was not able to
differentiate PD and HS.
The logistic LASSO model demonstrated that these altered local measures were clearly able to
distinguish the PD from the HS group. This implies that while individual ROIs may only be mildly
discriminative, collectively they provide a robust way to distinguish between groups. Although
not all ROIs selected by the logistic LASSO model were hubs, the results indicated an altered
network phenomenon may be initiated at hub regions.
Figure C.3 Logistic LASSO distinguishes PD and healthy subjects with all the local measures. In the analysis,
PD subjects were assigned 1 and healthy subjects were assigned 0 as their labels. Logistic LASSO predicts the
labels based on all local measures (both local efficiency and betweenness centrality). The estimated/predicted
labels are accurately categorized into two groups.
236
C.3 Canonical correlation analysis (CCA) between cognitive performance and global
measures
We further tested whether graphical measures were correlated with cognitive performance in a
multivariate manner using CCA. All global measures were concatenated to form one set and the
other set included cognitive scores. Likewise, the local measures which showed differences in the
t-tests were included as one set and cognitive scores were included as the other set. All variables
were normalized into z-scores and then processed with CCA. The canonical loadings with leave-
one-out cross validation of all variables were reported and a permutation test was carried out with
1000 iterations to evaluate the significance of the correlation in CCA.
The global graphical measures showed a significant correlation with behavioral scores in CCA
(r=0.98, p=0.01, Figure C.4). The variables were considered significant if the error bars did not
cross zero. In global measures, assortativity, modularity, rich club coefficient, and transitivity
demonstrated significant canonical loadings; while MoCA, BJLO, HVLY delay recall,
standardized HVLT total, standardized HVLT retention, LNS, and SF scores appeared influential
in the model. Modularity, rich club coefficient, and transitivity were positively correlated with
scores of BJLO, HVLY delay recall, standardized HVLT total, standardized HVLT retention, and
SF. In the other hand, assortativity was correlated with MoCA and LSN performance.
A CCA model attempting to relate local graphical measures and cognitive scores was not
significant.
237
Figure C.4 CCA reveals inter-correlations between global graphical measures and cognitive scores (correlation
= 0.98, p = 0.01, left panel). The error bars indicate 95% confidence interval in leave-one-out cross validation.
If the error bars do not cross zero, the variables are recognized as significant. Among global measures (middle
panel), assortativity, modularity and transitivity show significant loadings. In the combination of cognitive
scores (right panel), MoCA, BJLO, HVLY delay recall, standardized HVLT total, standardized HVLT
retention, LNS, and SF scores contribute significant loadings.
[assor: assortativity, chapath: characteristic path length, rich club: rich club coefficient, trans: transitivity,
MOCA: Montreal Cognitive Assessment, BJLOTOT = Bento Line Orientation Total Score, HVLTTOT =
Hopkins Verbal Learning Test-Revised Total Score, HVLTDELAY = HVLT Delayed Recall Score, DVT-
HVLTTOTAL = standardized HVLT Total Score, DVT-HVLTDELAY = standardized HVLT Delayed Recall
Score, DVT-HVLTRETENTION = standardized HVLT Recognition Trial Score, LNS = raw Letter-Number
Sequencing Test Score, SFCOM = Sematic Fluency Test – combination, SDMTTOT = Symbol Digit Modalities
Test total scores]
We found that segregation-oriented brain organization was related to better cognitive performance
with a multivariate approach. Modularity and transitivity, which are both measures of segregation,
were significantly correlated with better performances in line orientation, verbal learning, and
sematic fluency tests. This pattern indicated that better visuospatial, verbal learning and memory
238
functions, and executive skills were correlated with segregated brain organization in PD. In
addition, rich club coefficient, an indication of rich club structure which is thought to facilitate
cognitive processes [van den Heuvel and Sporns, 2011], was also related to better performance of
the above-mentioned tests, further emphasizing the important role of rich club structure in
cognition. On the other hand, assortativity, which shows the tendency of nodes with similar
connectivity to link with each other, was correlated with overall cognitive function as well as
memory, attention, and mental manipulation skills. However, as assortativity was ultimately a
correlation coefficient between node degrees and the measure was small in all subjects (maximum
value=0.09, minimum value=-0.1), we did not think this measure itself represented very
meaningful information in this cohort even though it was correlated with MoCA and LNS scores.
Therefore, taken together, we interpreted that multivariate approach also revealed similar findings
as univariate analysis but with greater robustness, whereby better cognitive function was related
to segregated brain networks as well as rich club characteristics in PD.
In this study, we did not report significant relations between local measures and cognitive
performance. This is somewhat surprising since the local graphical measures could be used to
distinguish between subject groups, while the global measures could not. The reason could be that
the overall, global brain organization is more related to cognitive performance, while smaller, more
subtle changes in local network characteristics may be related to other non-cognitive differences
between groups (e.g. related to motor function).
239
Appendix D
Appendix D lists the supporting materials of Chapter 5.
D.1 Linear regression model of dynamic features and WCSTCC
The Wisconsin Card Sorting Test Complete Categories score was modulated by one principal
component analysis (PCA) component of dynamic features as shown in the following table.
However, this component only explained 1.8% of the variance. In addition, this component was
dominated by the effects of flexibility of interhemispheric connections (FOCcs). The PCA
coefficient of FOCcs was 0.7 and the rest features showed coefficients ranged between -0.13 and
-0.36.
For individual predictors
behavioural
score
PCA components of
dynamic feature
estimate in regression
model
standard
error
p
values WCSTCC component 1 0.01 0.1 0.94
component 2 0.14 0.28 0.61
component 3 0.18 0.35 0.61
component 4 2.45 0.69 0.001
age (covariance) -0.32 0.02 0.15
For the whole model
Number of observations: 46, Error degrees of freedom: 40
Root Mean Squared Error: 1.5
R-squared: 0.314, Adjusted R-Squared 0.229
p-value = 0.008
[WCSTCC: the Wisconsin Card Sorting Test Complete Categories]
240
D.2 The component which significantly modulated cognitive flexibility
The flexibility of interhemispheric connections (FOCcs) loaded heavily than other features in
component 4. Although this component only explained limited variance of the data, it could predict
the performance of Wisconsin Card Sorting Test Complete Categories.